diff --git a/AGENTS.md b/AGENTS.md index 39183601..8609b02f 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -2,7 +2,7 @@ > **For:** Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Windsurf Cascade, and any AI coding agent. > **Repo:** `learning_ai_common_plat` — Shared platform packages + microservices for the ByteLyst ecosystem. -> **See also:** [`README.md`](README.md) for quick start, [`docs/ECOSYSTEM_ARCHITECTURE.md`](docs/ECOSYSTEM_ARCHITECTURE.md) for full architecture. +> **See also:** [`README.md`](README.md) for quick start, [`docs/ECOSYSTEM_ARCHITECTURE.md`](docs/ECOSYSTEM_ARCHITECTURE.md) for full architecture, [`docs/ecosystem/README.md`](docs/ecosystem/README.md) for canonical ecosystem strategy/contract docs. --- diff --git a/README.md b/README.md index bacb9147..11f2868f 100644 --- a/README.md +++ b/README.md @@ -53,6 +53,10 @@ Blob uploads are backed by local Azurite, prototype email delivery is backed by - **Dashboards** — `admin-web`, `tracker-web`, and `ux-lab`. - **MCP/A2A** — `services/mcp-server/` exposes tool routing, platform operator tools, extraction helpers, dev tools, product namespaces, and A2A orchestration pipelines. +## Ecosystem Docs + +Cross-product strategy and shared-contract documentation now lives under [`docs/ecosystem/`](/Users/saravana/BytelystAI/learning_ai/learning_ai_common_plat/docs/ecosystem). + ## Repository Structure ``` diff --git a/docs/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md b/docs/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md index e6f7d83c..414e855d 100644 --- a/docs/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md +++ b/docs/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md @@ -1,529 +1,7 @@ -# ByteLyst Ecosystem Cross-Pollination Opportunities +# Moved -> **Date:** 2026-04-03 -> **Scope:** Full local workspace scan across product repos, shared platform repos, internal tools, and `oss` agent repos. -> **Primary goal:** Identify concrete opportunities where one repo's capabilities can strengthen other repos or the shared platform. +This document now lives at: -## Design Principle +- [`docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md`](/Users/saravana/BytelystAI/learning_ai/learning_ai_common_plat/docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md) -The ecosystem should optimize for the fewest possible user steps. - -- Capture once. -- Understand once. -- Route automatically. -- Ask for approval only when risk or ambiguity justifies it. -- Let one action create value across multiple products. - -The more the ecosystem can turn one capture, one intent, or one approval into multiple useful downstream outcomes, the more defensible it becomes for users. - ---- - -## 1. Scan Coverage - -This review looked at the top-level README, `AGENTS.md`, `shared/product.json`, and top-level docs/layout signals for these repos: - -- `learning_ai_common_plat` -- `learning_voice_ai_agent` -- `learning_multimodal_memory_agents` -- `learning_ai_clock` -- `learning_ai_fastgap` -- `learning_ai_jarvis_jr` -- `learning_ai_peakpulse` -- `learning_ai_notes` -- `learning_ai_flowmonk` -- `learning_ai_trails` -- `learning_ai_local_memory_gpt` -- `learning_ai_local_llms` -- `learning_ai_smart_auth` -- `learning_ai_auth_app` -- `learning_ai_efforise` -- `learning_ai_productivity_web` -- `learning_ai_mac_tooling` -- `oss/learning_ai_claw-code-oss` -- `oss/learning_ai_claw-cowork` - -This is not a line-by-line code audit. It is a systematic ecosystem opportunity map grounded in the current repo surfaces that are already present in the workspace. - ---- - -## 2. Ecosystem Capability Map - -| Repo | Current Strength | Best Reusable Value for the Rest of the Ecosystem | -| ----------------------------------- | ---------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -| `learning_ai_common_plat` | Shared packages, platform-service, extraction-service, MCP/A2A, dashboards | The canonical place to centralize auth, telemetry, push, marketplace, usage, feature flags, design tokens, SDKs, and cross-product admin views | -| `learning_voice_ai_agent` | Cross-platform voice dictation, desktop-first workflows, transcripts, user/admin portals | Voice capture, transcript pipelines, desktop app patterns, billing and subscription patterns | -| `learning_multimodal_memory_agents` | Multimodal second-brain and memory primitives | Shared memory graph, reflection, summary, and personal knowledge patterns | -| `learning_ai_clock` | Timer/routine/focus orchestration across web/mobile/watch/mac | Time-awareness engine, routines, cascades, reminder orchestration | -| `learning_ai_fastgap` | Fasting visualization and coaching | Habit streaks, body-state workflows, longitudinal health tracking patterns | -| `learning_ai_jarvis_jr` | Voice-first multi-agent coaching and marketplace concepts | Agent marketplace, coaching orchestration, team/agent memory concepts | -| `learning_ai_peakpulse` | Sensor-driven adventure tracking | Sensor ingestion, route/session tracking, watch-first experiences | -| `learning_ai_notes` | Structured notes for humans and agents | Shared note blocks, capture flows, linking, retrieval, agent-readable context | -| `learning_ai_flowmonk` | Agent-first planning and execution | Planner UX, task decomposition, deterministic execution, structured work orchestration | -| `learning_ai_trails` | AI activity oversight, approvals, rewind, SDK | Agent audit, review, approval, replay, and observability primitives | -| `learning_ai_local_memory_gpt` | Local-first AI memory, RAG, multi-model compare | On-device memory, local vector/data patterns, privacy-first AI workflows | -| `learning_ai_local_llms` | Local model lab, hardware guidance, dashboard, eval workflows | Model selection, local inference guidance, hybrid-cloud AI fallback strategy | -| `learning_ai_smart_auth` | Identity platform spec | Canonical identity, passkey, MFA, OAuth, and org auth roadmap | -| `learning_ai_auth_app` | Standalone MFA/TOTP/device trust mobile app | Cross-product auth companion and secure approval UX | -| `learning_ai_efforise` | Identity-based habit tracking | Habit engine, nudges, streaks, effort scoring, lightweight wellness loops | -| `learning_ai_productivity_web` | Internal Next.js utilities shell | Fast internal operations tools, admin prototypes, workflow utilities | -| `learning_ai_mac_tooling` | Mac security audit, exfiltration detection, menu bar app | Device trust, data loss prevention, endpoint telemetry, secure desktop posture | -| `oss/learning_ai_claw-code-oss` | Open agent harness runtime | Sub-agents, todos/tasks, persistent project memory, session resume, tool permissioning, MCP lifecycle, config hierarchy, worktree isolation, remote trigger and bridge ideas | -| `oss/learning_ai_claw-cowork` | Desktop file automation agent with sandbox, plugins, connectors | Safe agent execution, scheduled automation, browser-extension capture, computer use, audit logs, prompt-injection detection, plugin marketplace, Dispatch API, MCP connectors | - ---- - -## 3. The Highest-Leverage Cross-Pollination Themes - -### 3.1 Identity, trust, and device posture - -The workspace has unusually strong raw ingredients for a serious ecosystem-grade trust layer: - -- `learning_ai_smart_auth` defines the auth direction -- `learning_ai_auth_app` provides the approval/MFA companion surface -- `learning_ai_common_plat` already hosts product-agnostic auth -- `learning_ai_mac_tooling` can contribute device trust and exfiltration signals -- `learning_ai_trails` can contribute agent approval and action replay -- `oss/learning_ai_claw-cowork` already has permissioned agent execution and budget controls - -This should converge into a single "trusted execution + trusted identity" spine instead of several parallel concepts. - -### 3.2 Memory, capture, and knowledge reuse - -There is a clear cluster around capture and persistent context: - -- `learning_voice_ai_agent` captures speech and transcripts -- `learning_ai_notes` stores structured notes -- `learning_multimodal_memory_agents` owns reflective memory and second-brain behaviors -- `learning_ai_local_memory_gpt` owns privacy-first local memory and RAG -- `learning_ai_jarvis_jr` and `learning_ai_flowmonk` both benefit from persistent agent context -- `oss/learning_ai_claw-cowork` could generate structured work artifacts into the same graph - -Right now this looks like multiple memory systems instead of one ecosystem knowledge substrate with different UX shells. - -### 3.3 Time, habits, goals, and execution loops - -A second cluster exists around time-aware action: - -- `learning_ai_clock` has routines, timers, and focus -- `learning_ai_efforise` has habit and effort loops -- `learning_ai_fastgap` tracks long-running behavior cycles -- `learning_ai_flowmonk` handles planning/execution -- `learning_ai_peakpulse` tracks activity sessions - -These should eventually share streaks, routines, reminders, schedules, and "goal progress" semantics. - -### 3.4 Agent platform and observability - -There is a strong but currently fragmented agent platform surface: - -- `learning_ai_trails` for oversight and approval -- `learning_ai_flowmonk` for planning and execution -- `learning_ai_jarvis_jr` for coaching agents and marketplace ideas -- `oss/learning_ai_claw-cowork` for sandboxed automation -- `oss/learning_ai_claw-code-oss` for harness/runtime patterns -- `learning_ai_common_plat` for MCP/A2A and marketplace infrastructure - -This is enough to build a coherent ByteLyst agent runtime and operations layer, but only if the contracts become shared. - -### 3.5 Low-friction personal operating system - -The biggest opportunity is not a larger feature list. It is collapsing multi-step workflows into one-step user outcomes. - -The `claw-code` and `claw-cowork` repos materially improve this because they already demonstrate reusable patterns for: - -- persistent sessions and resume -- todo/task tracking during long-running work -- remote triggers and scheduled triggers -- browser, desktop, and bridge handoff -- sub-agent or team-agent orchestration -- MCP-based tool routing and lifecycle -- explicit permission checkpoints -- plan, review, compact, and continue loops - -Those patterns should become ecosystem capabilities, not remain isolated CLI or desktop-agent behaviors. - ---- - -## 4. Ecosystem Gaps - -### 4.1 Shared platform gaps - -1. There is no single canonical cross-product user graph that combines identity, subscriptions, usage, device trust, approvals, memory artifacts, and activity history. -2. The workspace appears to have multiple "marketplace" concepts, but not one shared marketplace strategy spanning agents, plugins, prompts, skills, and premium workflows. -3. Cross-product notification orchestration is underleveraged. Several products need push, reminders, approvals, and escalations, but the experience is still product-local. -4. There is no clearly unified event taxonomy for actions like `capture.created`, `note.linked`, `agent.run.approved`, `timer.completed`, `habit.streak.updated`, `route.finished`, or `risk.device_elevated`. -5. Cross-product deep linking appears underdeveloped. The ecosystem needs one interoperable scheme so products can hand off context intentionally. -6. Several repos clearly rely on design tokens and shared platform patterns, but the ecosystem still looks uneven in auth, UI shell, token adoption, and package standardization. -7. There is no obvious shared "personal timeline" or "activity stream" that merges transcripts, notes, routines, sessions, tasks, trails, routes, and coaching moments. - -### 4.2 AI architecture gaps - -1. Memory is fragmented between cloud memory, local memory, notes, transcript history, and product-specific session stores. -2. There is no ecosystem-level retrieval layer that can pull from notes, transcripts, routines, plans, trails, and local/private stores under one permission model. -3. Model-routing strategy is not yet clearly shared across cloud-first, local-first, and hybrid products. -4. Agent execution policies are still product-specific instead of policy-driven and reusable across cowork, JarvisJr, FlowMonk, and ActionTrail. -5. Evaluation loops exist in pockets, but there is no single benchmark suite covering extraction quality, summarization quality, coaching quality, planning quality, and safety/approval precision. -6. There is no shared runtime contract for session memory, todo state, plan state, compaction, resume, worktree isolation, or remote triggers across agent products. -7. Browser extension capture, desktop Dispatch APIs, and bridge patterns exist in the `claw` repos, but they are not yet treated as reusable ecosystem primitives. - -### 4.3 Product experience gaps - -1. The ecosystem lacks clear handoff flows between products. Example: voice capture should become notes, tasks, memory, routines, and trails with almost no friction. -2. Wellness/time/planning products still read like separate apps rather than one coordinated life/work operating system. -3. Agent products do not yet visibly converge on one review model for approvals, rollback, trust, and audit. -4. Internal tools exist, but there is no single "operator cockpit" for ecosystem health, trust posture, cost, AI quality, and active incidents. -5. Browser-first and desktop-first flows are disconnected. The `Send to Cowork` style capture path should exist beyond file automation. -6. The ecosystem does not yet exploit resume-anywhere behavior. Users should be able to start on desktop, approve on mobile, inspect in trails, and consume outputs in notes or memory without restarting context. - ---- - -## 5. One-Step User Journeys To Optimize For - -These are the highest-value UX outcomes because they reduce user effort while increasing ecosystem leverage. - -1. **Speak once, benefit everywhere.** - A user dictates into LysnrAI and the system automatically creates a transcript, a structured NoteLett note, a MindLyst memory candidate, and optional FlowMonk tasks. - -2. **Plan once, follow through automatically.** - A user creates a plan in FlowMonk and it automatically becomes ChronoMind routines, EffoRise habits, and timeline entries, with approvals only when risky automation is involved. - -3. **Capture once from the browser.** - A user clicks a shared browser action and sends content into the right downstream product: NoteLett for knowledge, FlowMonk for action, MindLyst for memory, or Cowork for execution. - -4. **Delegate once, review once.** - A user dispatches a task to Cowork or another agent surface, receives approvals through Auth App when needed, reviews execution in ActionTrail, and gets the final artifact in NoteLett or MindLyst. - -5. **Start anywhere, resume anywhere.** - A user starts on desktop, receives a prompt or approval on mobile, revisits the audit trail on web, and resumes without reconstructing context. - -6. **One trusted approval, many safe actions.** - Once a device and user are trusted, the ecosystem should batch or suppress redundant approvals and only interrupt again when risk posture changes. - -These journeys should be used as the main acceptance test for cross-product architecture decisions. - ---- - -## 6. Cross-Pollination Opportunities by Capability Area - -### 6.1 Identity and security - -1. Use `learning_ai_auth_app` as the default approval device for ecosystem-wide login, sensitive action approval, and agent-run approval. -2. Bring `learning_ai_mac_tooling` device trust signals into `learning_ai_common_plat` so risky desktops can require step-up auth before agent execution or file export. -3. Reuse `learning_ai_trails` approval and rewind ideas inside `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, and `learning_ai_jarvis_jr`. -4. Implement a shared "trusted device score" service in `learning_ai_common_plat` backed by auth events, device posture, and anomaly signals. -5. Make SmartAuth the source of truth for passkeys, push MFA, device trust, and org policy enforcement across every product repo. -6. Reuse Cowork's destructive-action approvals, unattended execution policies, and audit-log patterns for higher-autonomy workflows across the ecosystem. - -### 6.2 Notes, memory, and knowledge graph - -1. Pipe LysnrAI transcripts directly into NoteLett note blocks and MindLyst memory ingestion. -2. Use NoteLett as the canonical human-editable layer on top of MindLyst's machine-generated memory/reflection layer. -3. Let Local Memory GPT sync selected artifacts from NoteLett and MindLyst into a local encrypted cache for offline/private recall. -4. Expose a common "artifact" schema in `learning_ai_common_plat` for transcript, note, summary, memory, plan, trail, route, and report objects. -5. Let Claw Cowork generate structured artifacts into the same schema after completing a task, so agent work products become searchable ecosystem knowledge. -6. Use `learning_ai_trails` to show provenance for who or what created each memory artifact, including user edit versus agent generation. -7. Reuse `claw-code` style persistent memory files and session-resume concepts for per-user, per-project, and per-goal continuity in FlowMonk, JarvisJr, and NoteLett. - -### 6.3 Voice and multimodal capture - -1. Reuse LysnrAI capture/transcript flows in JarvisJr, NoteLett, FlowMonk, and MindLyst instead of rebuilding voice capture independently. -2. Feed PeakPulse session voice notes and ChronoMind spoken commands into the same shared transcript/capture pipeline. -3. Add "capture once, route everywhere" workflows: one voice memo can become a note, task, habit check-in, route annotation, or memory entry. -4. Use extraction-service to normalize voice, document, screenshot, and scanned artifact ingestion for NoteLett, MindLyst, and Cowork. -5. Reuse Cowork browser-extension capture patterns for ecosystem-wide "Send to ..." flows from the browser into NoteLett, FlowMonk, MindLyst, and ActionTrail. - -### 6.4 Time, habits, planning, and coaching - -1. Reuse ChronoMind routines and reminder cascades inside EffoRise habit plans, NomGap fasting windows, and FlowMonk execution plans. -2. Let FlowMonk generate plans that can instantiate ChronoMind routines and EffoRise habits automatically. -3. Let JarvisJr coaching sessions schedule concrete follow-through actions in ChronoMind and EffoRise instead of ending as chat only. -4. Use PeakPulse activity sessions as signal input for EffoRise effort scoring and MindLyst daily reflection generation. -5. Add a shared "goal engine" in common platform that ties habits, plans, timers, sessions, and streaks into one measurable model. -6. Reuse `claw-code` todo/task patterns and Cowork scheduled automation patterns so plans can persist, resume, and self-advance with fewer user restarts. - -### 6.5 Agent runtime and oversight - -1. Converge on ActionTrail as the canonical review/audit layer for all higher-autonomy agent products. -2. Reuse Cowork's sandbox, plugin, connector, and budget-control ideas for JarvisJr and FlowMonk execution surfaces. -3. Use common platform MCP/A2A as the standard transport for inter-agent and product-to-tool orchestration instead of ad hoc connectors. -4. Define one ecosystem action log schema for planner decisions, tool calls, policy checks, approvals, retries, failures, and rollbacks. -5. Bring Claw Code OSS runtime learnings into a reusable internal harness package, but keep the ByteLyst policy/audit layer above it. -6. Make ActionTrail the replay UI for Cowork runs, FlowMonk automations, and future JarvisJr delegations. -7. Reuse `claw-code` features such as session resume, todo write, project memory, slash-command ergonomics, model/cost status, worktree isolation, and MCP lifecycle management as ecosystem runtime primitives rather than isolated CLI features. -8. Reuse Cowork's Dispatch API pattern so users can send a task from phone, browser, or another product into a trusted desktop executor without rebuilding orchestration from scratch. -9. Reuse Cowork's prompt-injection detection and screenshot manipulation detection for any product that lets agents inspect browser content, uploads, screenshots, or desktop surfaces. - -### 6.6 Marketplace, packaging, and monetization - -1. Unify the marketplace concepts in JarvisJr, Cowork, and common platform into one ecosystem marketplace for agents, plugins, templates, connectors, and premium packs. -2. Treat Note templates, FlowMonk workflows, ChronoMind routines, EffoRise programs, and Cowork skills as marketplace inventory types rather than isolated config files. -3. Reuse platform-service billing, entitlement, and usage controls for premium automations and multi-product bundles. -4. Support "creator packs" that can publish one asset to multiple products with product-specific presentation layers. -5. Treat skills, prompt packs, review templates, slash-command bundles, and department agents from the `claw` repos as first-class marketplace inventory candidates. - -### 6.7 Notifications and cross-device experiences - -1. Use a shared notification orchestration layer for MFA approval, timer urgency, coaching nudges, habit reminders, plan deadlines, and agent approval prompts. -2. Let ByteLyst Auth become the secure approval channel, while ChronoMind and EffoRise become the routine/nudge channels. -3. Feed watch/mobile surfaces from ChronoMind, PeakPulse, and Auth into a coordinated wearable strategy instead of separate notification silos. -4. Build shared digest surfaces: daily brief, risk brief, work brief, health brief, and memory brief. -5. Add "resume on another surface" interactions inspired by `claw-code` and Cowork: start in desktop, continue in web, approve in auth app, inspect in trails, consume in notes or memory. - -### 6.8 Internal operations and developer tooling - -1. Fold `learning_ai_productivity_web` into a general-purpose internal operations shell or replace it with shared platform operator dashboards. -2. Expose `learning_ai_mac_tooling` findings into admin-web so security posture becomes visible next to auth, usage, and incident data. -3. Use local-llm-lab as the canonical place for model benchmarking before products adopt new local or hybrid model defaults. -4. Add a workspace-wide capability registry document generated from product manifests and package usage, not maintained manually. - ---- - -## 7. Improvement Ideas by Repo - -### `learning_ai_common_plat` - -- Add a cross-product artifact schema and artifact index service. -- Add a shared event taxonomy and event explorer dashboard. -- Add a trusted-device / device-risk service. -- Add a unified marketplace inventory model for templates, plugins, prompts, skills, agents, and workflows. -- Add a personal timeline API that aggregates cross-product activity. -- Add cross-product notification orchestration with channel rules and escalation rules. -- Add a reusable agent runtime contract covering session state, todos, plan state, resume, approvals, action logs, and handoff metadata. -- Add a bridge/dispatch abstraction for browser, IDE, mobile, and desktop task handoff. - -### `learning_voice_ai_agent` - -- Become the canonical capture service for voice-first input across the ecosystem. -- Export transcripts directly into NoteLett, MindLyst, and FlowMonk. -- Reuse ActionTrail to show transcript edits, exports, and agent transformations. - -### `learning_multimodal_memory_agents` - -- Become the canonical reflection and memory synthesis layer for notes, transcripts, habits, sessions, and routes. -- Accept imports from Cowork reports, PeakPulse sessions, and ChronoMind routines. -- Expose memory APIs consumable by JarvisJr and Local Memory GPT. - -### `learning_ai_clock` - -- Publish routines/reminder APIs that EffoRise, FlowMonk, JarvisJr, and NomGap can call. -- Share urgency and escalation models with auth approvals and agent approvals. -- Become the schedule engine for cross-product follow-through. - -### `learning_ai_fastgap` - -- Reuse EffoRise habits and ChronoMind routines rather than owning separate streak/reminder logic. -- Feed fasting state into coaching and reflection systems. -- Export milestone events into the shared personal timeline. - -### `learning_ai_jarvis_jr` - -- Reuse common platform marketplace instead of a product-local marketplace. -- Consume MindLyst memory and NoteLett notes as coaching context. -- Send delegated or automated actions through ActionTrail and Cowork-style policy gates. -- Reuse `claw-code` team/sub-agent, task, and persistent-memory patterns for coach specialization and delegated assistance. - -### `learning_ai_peakpulse` - -- Reuse MindLyst for post-activity reflection and journaling. -- Reuse ChronoMind for pacing alerts and preparation warnings. -- Feed route/session output into EffoRise effort scoring and shared activity timelines. - -### `learning_ai_notes` - -- Become the canonical editable workspace for artifacts generated by LysnrAI, Cowork, FlowMonk, and JarvisJr. -- Add first-class provenance and "created by agent" metadata from ActionTrail. -- Share block schemas with MindLyst and Local Memory GPT. -- Add browser and desktop quick-capture entrypoints modeled after Cowork's extension and dispatch patterns. - -### `learning_ai_flowmonk` - -- Build on ChronoMind routines, EffoRise habits, and NoteLett notes instead of reproducing those abstractions. -- Use Cowork for high-autonomy execution and ActionTrail for audit/approval. -- Export plans as artifacts and timeline events into common platform. -- Reuse `claw-code` todo lists, plan mode, compaction, and resume semantics so plans can survive long-running work without context loss. - -### `learning_ai_trails` - -- Become the ecosystem action ledger for every meaningful AI-driven operation. -- Add first-class adapters for Cowork, FlowMonk, JarvisJr, and LysnrAI transformations. -- Integrate device trust and auth state into approval policy decisions. -- Add replay views for browser capture, remote dispatch, scheduled runs, and resume chains across agent workflows. - -### `learning_ai_local_memory_gpt` - -- Serve as the privacy-preserving local cache and compare UI for selected ecosystem knowledge. -- Sync approved notes, transcripts, and summaries from cloud products. -- Reuse local-llm-lab model guidance and common platform artifact schemas. - -### `learning_ai_local_llms` - -- Become the standard evaluation and recommendation lab for local/hybrid model adoption across products. -- Publish a machine-readable capability matrix for models and hardware. -- Feed recommendation outputs into Local Memory GPT, Cowork, and any offline-capable products. - -### `learning_ai_smart_auth` - -- Move from spec-only posture toward a concrete reference architecture doc that maps directly to platform-service modules, auth-app features, and device trust integrations. - -### `learning_ai_auth_app` - -- Expand beyond login MFA into ecosystem action approval, budget escalation approval, and org admin confirmation. -- Add explicit support for approving agent runs and sensitive exports. - -### `learning_ai_efforise` - -- Reuse ChronoMind schedules, PeakPulse activity data, and MindLyst reflections to make habits feel contextual rather than isolated. - -### `learning_ai_productivity_web` - -- Either formalize as the lightweight internal ops sandbox or retire in favor of common platform dashboards and UX lab. -- If kept, use it as a thin shell over shared bridge, dispatch, and operator APIs rather than a separate island. - -### `learning_ai_mac_tooling` - -- Integrate with SmartAuth and platform-service as a device risk source. -- Feed exfiltration alerts into ActionTrail and admin-web. -- Use Cowork policies to limit high-risk local automation when the device is degraded. - -### `oss/learning_ai_claw-code-oss` - -- Reuse only the clean runtime/harness ideas that strengthen internal agent execution: session persistence, task/todo state, project memory, slash-command ergonomics, MCP lifecycle, config hierarchy, bridge modes, remote triggers, and worktree isolation. -- Do not allow the workspace to drift into a second ungoverned policy layer. - -### `oss/learning_ai_claw-cowork` - -- Integrate tightly with ActionTrail, SmartAuth, Auth App, and common platform marketplace. -- Emit structured artifacts into NoteLett and MindLyst after runs. -- Reuse local-llm-lab recommendations for local execution fallback and cost control. -- Promote its strongest reusable capabilities into the ecosystem: Dispatch API, scheduled automation, browser extension capture, audit panel patterns, prompt-injection detection, screenshot safety heuristics, and app/browser/desktop handoff. - ---- - -## 8. Priority Recommendations - -### P0: Foundational integrations - -1. Define a shared artifact schema in `learning_ai_common_plat`. -2. Define a shared event taxonomy and action log schema. -3. Define a unified approval model spanning SmartAuth, Auth App, ActionTrail, and Cowork. -4. Define a shared cross-product deep-linking convention. -5. Define a marketplace inventory model that supports plugins, agents, prompts, templates, and workflows. -6. Define a shared agent runtime contract for session memory, todo state, approvals, resume, dispatch, and schedule metadata. - -### P1: Cross-product user value - -1. LysnrAI transcript -> NoteLett note -> MindLyst memory pipeline. -2. FlowMonk plan -> ChronoMind routine -> EffoRise habit handoff. -3. Cowork run -> ActionTrail audit -> NoteLett report -> MindLyst memory ingestion. -4. PeakPulse session -> MindLyst reflection -> EffoRise effort credit. -5. Device risk from mac-tooling -> SmartAuth step-up auth -> Cowork execution restrictions. -6. Browser/web capture -> NoteLett or FlowMonk or MindLyst via a shared "Send to ..." entrypoint. -7. Remote task dispatch -> trusted desktop executor -> trails replay -> notes or memory output. - -### P2: Ecosystem differentiation - -1. Shared personal timeline across all products. -2. Shared notification and digest orchestration. -3. Shared creator marketplace across workflows, plugins, notes, routines, and coaching packs. -4. Hybrid cloud/local memory and model-routing layer combining common platform, Local LLM Lab, and Local Memory GPT. - ---- - -## 9. Execution Matrix - -| Initiative | Primary Owner | Key Dependent Repos | Impact | Effort | Why It Should Happen Early | -| ----------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ------ | ------ | ----------------------------------------------------------------------------------------------------- | -| Shared artifact schema | `learning_ai_common_plat` | `learning_voice_ai_agent`, `learning_ai_notes`, `learning_multimodal_memory_agents`, `learning_ai_flowmonk`, `oss/learning_ai_claw-cowork` | High | Medium | It unblocks memory, search, provenance, timelines, and agent output reuse | -| Shared event taxonomy + action log | `learning_ai_common_plat` + `learning_ai_trails` | `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, `learning_ai_jarvis_jr`, `learning_voice_ai_agent` | High | Medium | It turns fragmented agent/product telemetry into one reviewable substrate | -| Unified approvals and trust model | `learning_ai_smart_auth` + `learning_ai_auth_app` + `learning_ai_trails` | `oss/learning_ai_claw-cowork`, `learning_ai_mac_tooling`, `learning_ai_common_plat` | High | Medium | It creates one policy model for login, step-up auth, and agent approvals | -| Shared agent runtime contract | `learning_ai_common_plat` | `oss/learning_ai_claw-code-oss`, `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, `learning_ai_jarvis_jr`, `learning_ai_trails` | High | Medium | It prevents each agent product from reinventing session state, todos, approvals, and resume flows | -| Transcript -> note -> memory pipeline | `learning_voice_ai_agent` + `learning_ai_notes` + `learning_multimodal_memory_agents` | `learning_ai_common_plat` | High | Medium | It creates immediate user-visible value across three flagship products | -| Plan -> routine -> habit handoff | `learning_ai_flowmonk` + `learning_ai_clock` + `learning_ai_efforise` | `learning_ai_common_plat` | High | Medium | It turns planning into follow-through instead of isolated planning data | -| Cowork -> Trail -> Notes -> Memory flow | `oss/learning_ai_claw-cowork` + `learning_ai_trails` + `learning_ai_notes` + `learning_multimodal_memory_agents` | `learning_ai_common_plat` | High | High | It creates a differentiated audited-agent workflow that few ecosystems have | -| Device trust ingestion | `learning_ai_mac_tooling` + `learning_ai_common_plat` | `learning_ai_smart_auth`, `learning_ai_auth_app`, `oss/learning_ai_claw-cowork` | Medium | Medium | It upgrades desktop/agent safety with existing assets already in the workspace | -| Shared browser and desktop dispatch layer | `learning_ai_common_plat` + `oss/learning_ai_claw-cowork` | `learning_ai_notes`, `learning_ai_flowmonk`, `learning_multimodal_memory_agents`, `learning_ai_trails` | High | Medium | It lowers capture friction and creates one route from browser or mobile into agent-assisted workflows | -| Shared marketplace inventory model | `learning_ai_common_plat` | `learning_ai_jarvis_jr`, `oss/learning_ai_claw-cowork`, `learning_ai_notes`, `learning_ai_flowmonk`, `learning_ai_clock` | High | High | It can unify monetization and reusable asset distribution | -| Shared personal timeline | `learning_ai_common_plat` | Almost all product repos | High | High | It becomes the ecosystem shell once events and artifacts are standardized | -| Hybrid local/cloud AI routing | `learning_ai_local_llms` + `learning_ai_local_memory_gpt` + `learning_ai_common_plat` | `oss/learning_ai_claw-cowork`, `learning_multimodal_memory_agents`, `learning_voice_ai_agent` | Medium | High | It is strategically valuable, but depends on artifact and policy standardization first | - ---- - -## 10. Guardrails: What Not to Over-Unify - -Cross-pollination only helps if it centralizes the right seams. These areas should stay product-specific: - -1. Product voice, branding, and UX metaphors should remain local to each repo. -2. Domain logic should remain local when it is the core product moat. Examples: fasting logic in `learning_ai_fastgap`, adventure session semantics in `learning_ai_peakpulse`, timer cascades in `learning_ai_clock`. -3. Local-first privacy behavior in `learning_ai_local_memory_gpt` should not be weakened just to mirror cloud-first products. -4. Agent runtime experimentation in `oss/learning_ai_claw-code-oss` and `oss/learning_ai_claw-cowork` should stay decoupled from platform policy enough to evolve safely. -5. Shared platform should own contracts, policy, schemas, and orchestration primitives, but not erase the reason each product exists. - -The unification target should be: - -- identity -- approvals -- events -- artifacts -- runtime contracts -- dispatch and bridge contracts -- notifications -- marketplace inventory -- retrieval contracts -- observability - -The non-target should be each app's core domain experience. - ---- - -## 11. Suggested 90-Day Sequence - -### Days 1-30 - -1. Draft the shared artifact schema. -2. Draft the shared event taxonomy and action log schema. -3. Draft the unified approval and trusted-device model. -4. Draft the cross-product deep-link convention. -5. Draft the shared agent runtime contract for session state, resume, todos, schedules, and dispatch. - -### Days 31-60 - -1. Implement the LysnrAI -> NoteLett -> MindLyst pipeline. -2. Implement the FlowMonk -> ChronoMind -> EffoRise handoff. -3. Implement the Cowork -> ActionTrail audit event emission path. -4. Prototype mac-tooling -> platform-service device risk ingestion. -5. Prototype shared browser capture and dispatch using Cowork extension and Dispatch API patterns. - -### Days 61-90 - -1. Stand up a personal timeline API and internal timeline viewer. -2. Stand up a shared notification orchestration service. -3. Convert marketplace concepts into one inventory model and one entitlement model. -4. Add one operator cockpit view that merges auth, usage, risk, and AI action streams. -5. Add resume-anywhere flows across desktop, auth app, trails, and notes or memory consumers. - ---- - -## 12. Recommended Next Documents - -This document should lead to a small set of follow-on specs rather than another broad audit: - -1. `docs/ECOSYSTEM_SHARED_ARTIFACT_SCHEMA.md` -2. `docs/ECOSYSTEM_EVENT_TAXONOMY.md` -3. `docs/ECOSYSTEM_APPROVALS_AND_TRUST_MODEL.md` -4. `docs/ECOSYSTEM_MARKETPLACE_UNIFICATION.md` -5. `docs/ECOSYSTEM_PERSONAL_TIMELINE_PRD.md` - ---- - -## 13. Bottom Line - -The ecosystem's biggest unrealized strength is not any single app. It is the combination of: - -- strong identity and approvals -- strong voice and multimodal capture -- strong memory and note systems -- strong time/habit/planning loops -- strong agent execution and audit infrastructure -- strong harness and runtime patterns already proven in `claw-code` and `claw-cowork` - -The main gap is that these strengths still live in adjacent repos instead of one intentional platform model. The next step should be standardizing the shared contracts that let each repo keep its own product identity while participating in one ByteLyst operating system. +Use the `docs/ecosystem/` folder as the canonical home for ecosystem-level strategy, contracts, rollout plans, and cross-repo adoption docs. diff --git a/docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md b/docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md new file mode 100644 index 00000000..e6f7d83c --- /dev/null +++ b/docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md @@ -0,0 +1,529 @@ +# ByteLyst Ecosystem Cross-Pollination Opportunities + +> **Date:** 2026-04-03 +> **Scope:** Full local workspace scan across product repos, shared platform repos, internal tools, and `oss` agent repos. +> **Primary goal:** Identify concrete opportunities where one repo's capabilities can strengthen other repos or the shared platform. + +## Design Principle + +The ecosystem should optimize for the fewest possible user steps. + +- Capture once. +- Understand once. +- Route automatically. +- Ask for approval only when risk or ambiguity justifies it. +- Let one action create value across multiple products. + +The more the ecosystem can turn one capture, one intent, or one approval into multiple useful downstream outcomes, the more defensible it becomes for users. + +--- + +## 1. Scan Coverage + +This review looked at the top-level README, `AGENTS.md`, `shared/product.json`, and top-level docs/layout signals for these repos: + +- `learning_ai_common_plat` +- `learning_voice_ai_agent` +- `learning_multimodal_memory_agents` +- `learning_ai_clock` +- `learning_ai_fastgap` +- `learning_ai_jarvis_jr` +- `learning_ai_peakpulse` +- `learning_ai_notes` +- `learning_ai_flowmonk` +- `learning_ai_trails` +- `learning_ai_local_memory_gpt` +- `learning_ai_local_llms` +- `learning_ai_smart_auth` +- `learning_ai_auth_app` +- `learning_ai_efforise` +- `learning_ai_productivity_web` +- `learning_ai_mac_tooling` +- `oss/learning_ai_claw-code-oss` +- `oss/learning_ai_claw-cowork` + +This is not a line-by-line code audit. It is a systematic ecosystem opportunity map grounded in the current repo surfaces that are already present in the workspace. + +--- + +## 2. Ecosystem Capability Map + +| Repo | Current Strength | Best Reusable Value for the Rest of the Ecosystem | +| ----------------------------------- | ---------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| `learning_ai_common_plat` | Shared packages, platform-service, extraction-service, MCP/A2A, dashboards | The canonical place to centralize auth, telemetry, push, marketplace, usage, feature flags, design tokens, SDKs, and cross-product admin views | +| `learning_voice_ai_agent` | Cross-platform voice dictation, desktop-first workflows, transcripts, user/admin portals | Voice capture, transcript pipelines, desktop app patterns, billing and subscription patterns | +| `learning_multimodal_memory_agents` | Multimodal second-brain and memory primitives | Shared memory graph, reflection, summary, and personal knowledge patterns | +| `learning_ai_clock` | Timer/routine/focus orchestration across web/mobile/watch/mac | Time-awareness engine, routines, cascades, reminder orchestration | +| `learning_ai_fastgap` | Fasting visualization and coaching | Habit streaks, body-state workflows, longitudinal health tracking patterns | +| `learning_ai_jarvis_jr` | Voice-first multi-agent coaching and marketplace concepts | Agent marketplace, coaching orchestration, team/agent memory concepts | +| `learning_ai_peakpulse` | Sensor-driven adventure tracking | Sensor ingestion, route/session tracking, watch-first experiences | +| `learning_ai_notes` | Structured notes for humans and agents | Shared note blocks, capture flows, linking, retrieval, agent-readable context | +| `learning_ai_flowmonk` | Agent-first planning and execution | Planner UX, task decomposition, deterministic execution, structured work orchestration | +| `learning_ai_trails` | AI activity oversight, approvals, rewind, SDK | Agent audit, review, approval, replay, and observability primitives | +| `learning_ai_local_memory_gpt` | Local-first AI memory, RAG, multi-model compare | On-device memory, local vector/data patterns, privacy-first AI workflows | +| `learning_ai_local_llms` | Local model lab, hardware guidance, dashboard, eval workflows | Model selection, local inference guidance, hybrid-cloud AI fallback strategy | +| `learning_ai_smart_auth` | Identity platform spec | Canonical identity, passkey, MFA, OAuth, and org auth roadmap | +| `learning_ai_auth_app` | Standalone MFA/TOTP/device trust mobile app | Cross-product auth companion and secure approval UX | +| `learning_ai_efforise` | Identity-based habit tracking | Habit engine, nudges, streaks, effort scoring, lightweight wellness loops | +| `learning_ai_productivity_web` | Internal Next.js utilities shell | Fast internal operations tools, admin prototypes, workflow utilities | +| `learning_ai_mac_tooling` | Mac security audit, exfiltration detection, menu bar app | Device trust, data loss prevention, endpoint telemetry, secure desktop posture | +| `oss/learning_ai_claw-code-oss` | Open agent harness runtime | Sub-agents, todos/tasks, persistent project memory, session resume, tool permissioning, MCP lifecycle, config hierarchy, worktree isolation, remote trigger and bridge ideas | +| `oss/learning_ai_claw-cowork` | Desktop file automation agent with sandbox, plugins, connectors | Safe agent execution, scheduled automation, browser-extension capture, computer use, audit logs, prompt-injection detection, plugin marketplace, Dispatch API, MCP connectors | + +--- + +## 3. The Highest-Leverage Cross-Pollination Themes + +### 3.1 Identity, trust, and device posture + +The workspace has unusually strong raw ingredients for a serious ecosystem-grade trust layer: + +- `learning_ai_smart_auth` defines the auth direction +- `learning_ai_auth_app` provides the approval/MFA companion surface +- `learning_ai_common_plat` already hosts product-agnostic auth +- `learning_ai_mac_tooling` can contribute device trust and exfiltration signals +- `learning_ai_trails` can contribute agent approval and action replay +- `oss/learning_ai_claw-cowork` already has permissioned agent execution and budget controls + +This should converge into a single "trusted execution + trusted identity" spine instead of several parallel concepts. + +### 3.2 Memory, capture, and knowledge reuse + +There is a clear cluster around capture and persistent context: + +- `learning_voice_ai_agent` captures speech and transcripts +- `learning_ai_notes` stores structured notes +- `learning_multimodal_memory_agents` owns reflective memory and second-brain behaviors +- `learning_ai_local_memory_gpt` owns privacy-first local memory and RAG +- `learning_ai_jarvis_jr` and `learning_ai_flowmonk` both benefit from persistent agent context +- `oss/learning_ai_claw-cowork` could generate structured work artifacts into the same graph + +Right now this looks like multiple memory systems instead of one ecosystem knowledge substrate with different UX shells. + +### 3.3 Time, habits, goals, and execution loops + +A second cluster exists around time-aware action: + +- `learning_ai_clock` has routines, timers, and focus +- `learning_ai_efforise` has habit and effort loops +- `learning_ai_fastgap` tracks long-running behavior cycles +- `learning_ai_flowmonk` handles planning/execution +- `learning_ai_peakpulse` tracks activity sessions + +These should eventually share streaks, routines, reminders, schedules, and "goal progress" semantics. + +### 3.4 Agent platform and observability + +There is a strong but currently fragmented agent platform surface: + +- `learning_ai_trails` for oversight and approval +- `learning_ai_flowmonk` for planning and execution +- `learning_ai_jarvis_jr` for coaching agents and marketplace ideas +- `oss/learning_ai_claw-cowork` for sandboxed automation +- `oss/learning_ai_claw-code-oss` for harness/runtime patterns +- `learning_ai_common_plat` for MCP/A2A and marketplace infrastructure + +This is enough to build a coherent ByteLyst agent runtime and operations layer, but only if the contracts become shared. + +### 3.5 Low-friction personal operating system + +The biggest opportunity is not a larger feature list. It is collapsing multi-step workflows into one-step user outcomes. + +The `claw-code` and `claw-cowork` repos materially improve this because they already demonstrate reusable patterns for: + +- persistent sessions and resume +- todo/task tracking during long-running work +- remote triggers and scheduled triggers +- browser, desktop, and bridge handoff +- sub-agent or team-agent orchestration +- MCP-based tool routing and lifecycle +- explicit permission checkpoints +- plan, review, compact, and continue loops + +Those patterns should become ecosystem capabilities, not remain isolated CLI or desktop-agent behaviors. + +--- + +## 4. Ecosystem Gaps + +### 4.1 Shared platform gaps + +1. There is no single canonical cross-product user graph that combines identity, subscriptions, usage, device trust, approvals, memory artifacts, and activity history. +2. The workspace appears to have multiple "marketplace" concepts, but not one shared marketplace strategy spanning agents, plugins, prompts, skills, and premium workflows. +3. Cross-product notification orchestration is underleveraged. Several products need push, reminders, approvals, and escalations, but the experience is still product-local. +4. There is no clearly unified event taxonomy for actions like `capture.created`, `note.linked`, `agent.run.approved`, `timer.completed`, `habit.streak.updated`, `route.finished`, or `risk.device_elevated`. +5. Cross-product deep linking appears underdeveloped. The ecosystem needs one interoperable scheme so products can hand off context intentionally. +6. Several repos clearly rely on design tokens and shared platform patterns, but the ecosystem still looks uneven in auth, UI shell, token adoption, and package standardization. +7. There is no obvious shared "personal timeline" or "activity stream" that merges transcripts, notes, routines, sessions, tasks, trails, routes, and coaching moments. + +### 4.2 AI architecture gaps + +1. Memory is fragmented between cloud memory, local memory, notes, transcript history, and product-specific session stores. +2. There is no ecosystem-level retrieval layer that can pull from notes, transcripts, routines, plans, trails, and local/private stores under one permission model. +3. Model-routing strategy is not yet clearly shared across cloud-first, local-first, and hybrid products. +4. Agent execution policies are still product-specific instead of policy-driven and reusable across cowork, JarvisJr, FlowMonk, and ActionTrail. +5. Evaluation loops exist in pockets, but there is no single benchmark suite covering extraction quality, summarization quality, coaching quality, planning quality, and safety/approval precision. +6. There is no shared runtime contract for session memory, todo state, plan state, compaction, resume, worktree isolation, or remote triggers across agent products. +7. Browser extension capture, desktop Dispatch APIs, and bridge patterns exist in the `claw` repos, but they are not yet treated as reusable ecosystem primitives. + +### 4.3 Product experience gaps + +1. The ecosystem lacks clear handoff flows between products. Example: voice capture should become notes, tasks, memory, routines, and trails with almost no friction. +2. Wellness/time/planning products still read like separate apps rather than one coordinated life/work operating system. +3. Agent products do not yet visibly converge on one review model for approvals, rollback, trust, and audit. +4. Internal tools exist, but there is no single "operator cockpit" for ecosystem health, trust posture, cost, AI quality, and active incidents. +5. Browser-first and desktop-first flows are disconnected. The `Send to Cowork` style capture path should exist beyond file automation. +6. The ecosystem does not yet exploit resume-anywhere behavior. Users should be able to start on desktop, approve on mobile, inspect in trails, and consume outputs in notes or memory without restarting context. + +--- + +## 5. One-Step User Journeys To Optimize For + +These are the highest-value UX outcomes because they reduce user effort while increasing ecosystem leverage. + +1. **Speak once, benefit everywhere.** + A user dictates into LysnrAI and the system automatically creates a transcript, a structured NoteLett note, a MindLyst memory candidate, and optional FlowMonk tasks. + +2. **Plan once, follow through automatically.** + A user creates a plan in FlowMonk and it automatically becomes ChronoMind routines, EffoRise habits, and timeline entries, with approvals only when risky automation is involved. + +3. **Capture once from the browser.** + A user clicks a shared browser action and sends content into the right downstream product: NoteLett for knowledge, FlowMonk for action, MindLyst for memory, or Cowork for execution. + +4. **Delegate once, review once.** + A user dispatches a task to Cowork or another agent surface, receives approvals through Auth App when needed, reviews execution in ActionTrail, and gets the final artifact in NoteLett or MindLyst. + +5. **Start anywhere, resume anywhere.** + A user starts on desktop, receives a prompt or approval on mobile, revisits the audit trail on web, and resumes without reconstructing context. + +6. **One trusted approval, many safe actions.** + Once a device and user are trusted, the ecosystem should batch or suppress redundant approvals and only interrupt again when risk posture changes. + +These journeys should be used as the main acceptance test for cross-product architecture decisions. + +--- + +## 6. Cross-Pollination Opportunities by Capability Area + +### 6.1 Identity and security + +1. Use `learning_ai_auth_app` as the default approval device for ecosystem-wide login, sensitive action approval, and agent-run approval. +2. Bring `learning_ai_mac_tooling` device trust signals into `learning_ai_common_plat` so risky desktops can require step-up auth before agent execution or file export. +3. Reuse `learning_ai_trails` approval and rewind ideas inside `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, and `learning_ai_jarvis_jr`. +4. Implement a shared "trusted device score" service in `learning_ai_common_plat` backed by auth events, device posture, and anomaly signals. +5. Make SmartAuth the source of truth for passkeys, push MFA, device trust, and org policy enforcement across every product repo. +6. Reuse Cowork's destructive-action approvals, unattended execution policies, and audit-log patterns for higher-autonomy workflows across the ecosystem. + +### 6.2 Notes, memory, and knowledge graph + +1. Pipe LysnrAI transcripts directly into NoteLett note blocks and MindLyst memory ingestion. +2. Use NoteLett as the canonical human-editable layer on top of MindLyst's machine-generated memory/reflection layer. +3. Let Local Memory GPT sync selected artifacts from NoteLett and MindLyst into a local encrypted cache for offline/private recall. +4. Expose a common "artifact" schema in `learning_ai_common_plat` for transcript, note, summary, memory, plan, trail, route, and report objects. +5. Let Claw Cowork generate structured artifacts into the same schema after completing a task, so agent work products become searchable ecosystem knowledge. +6. Use `learning_ai_trails` to show provenance for who or what created each memory artifact, including user edit versus agent generation. +7. Reuse `claw-code` style persistent memory files and session-resume concepts for per-user, per-project, and per-goal continuity in FlowMonk, JarvisJr, and NoteLett. + +### 6.3 Voice and multimodal capture + +1. Reuse LysnrAI capture/transcript flows in JarvisJr, NoteLett, FlowMonk, and MindLyst instead of rebuilding voice capture independently. +2. Feed PeakPulse session voice notes and ChronoMind spoken commands into the same shared transcript/capture pipeline. +3. Add "capture once, route everywhere" workflows: one voice memo can become a note, task, habit check-in, route annotation, or memory entry. +4. Use extraction-service to normalize voice, document, screenshot, and scanned artifact ingestion for NoteLett, MindLyst, and Cowork. +5. Reuse Cowork browser-extension capture patterns for ecosystem-wide "Send to ..." flows from the browser into NoteLett, FlowMonk, MindLyst, and ActionTrail. + +### 6.4 Time, habits, planning, and coaching + +1. Reuse ChronoMind routines and reminder cascades inside EffoRise habit plans, NomGap fasting windows, and FlowMonk execution plans. +2. Let FlowMonk generate plans that can instantiate ChronoMind routines and EffoRise habits automatically. +3. Let JarvisJr coaching sessions schedule concrete follow-through actions in ChronoMind and EffoRise instead of ending as chat only. +4. Use PeakPulse activity sessions as signal input for EffoRise effort scoring and MindLyst daily reflection generation. +5. Add a shared "goal engine" in common platform that ties habits, plans, timers, sessions, and streaks into one measurable model. +6. Reuse `claw-code` todo/task patterns and Cowork scheduled automation patterns so plans can persist, resume, and self-advance with fewer user restarts. + +### 6.5 Agent runtime and oversight + +1. Converge on ActionTrail as the canonical review/audit layer for all higher-autonomy agent products. +2. Reuse Cowork's sandbox, plugin, connector, and budget-control ideas for JarvisJr and FlowMonk execution surfaces. +3. Use common platform MCP/A2A as the standard transport for inter-agent and product-to-tool orchestration instead of ad hoc connectors. +4. Define one ecosystem action log schema for planner decisions, tool calls, policy checks, approvals, retries, failures, and rollbacks. +5. Bring Claw Code OSS runtime learnings into a reusable internal harness package, but keep the ByteLyst policy/audit layer above it. +6. Make ActionTrail the replay UI for Cowork runs, FlowMonk automations, and future JarvisJr delegations. +7. Reuse `claw-code` features such as session resume, todo write, project memory, slash-command ergonomics, model/cost status, worktree isolation, and MCP lifecycle management as ecosystem runtime primitives rather than isolated CLI features. +8. Reuse Cowork's Dispatch API pattern so users can send a task from phone, browser, or another product into a trusted desktop executor without rebuilding orchestration from scratch. +9. Reuse Cowork's prompt-injection detection and screenshot manipulation detection for any product that lets agents inspect browser content, uploads, screenshots, or desktop surfaces. + +### 6.6 Marketplace, packaging, and monetization + +1. Unify the marketplace concepts in JarvisJr, Cowork, and common platform into one ecosystem marketplace for agents, plugins, templates, connectors, and premium packs. +2. Treat Note templates, FlowMonk workflows, ChronoMind routines, EffoRise programs, and Cowork skills as marketplace inventory types rather than isolated config files. +3. Reuse platform-service billing, entitlement, and usage controls for premium automations and multi-product bundles. +4. Support "creator packs" that can publish one asset to multiple products with product-specific presentation layers. +5. Treat skills, prompt packs, review templates, slash-command bundles, and department agents from the `claw` repos as first-class marketplace inventory candidates. + +### 6.7 Notifications and cross-device experiences + +1. Use a shared notification orchestration layer for MFA approval, timer urgency, coaching nudges, habit reminders, plan deadlines, and agent approval prompts. +2. Let ByteLyst Auth become the secure approval channel, while ChronoMind and EffoRise become the routine/nudge channels. +3. Feed watch/mobile surfaces from ChronoMind, PeakPulse, and Auth into a coordinated wearable strategy instead of separate notification silos. +4. Build shared digest surfaces: daily brief, risk brief, work brief, health brief, and memory brief. +5. Add "resume on another surface" interactions inspired by `claw-code` and Cowork: start in desktop, continue in web, approve in auth app, inspect in trails, consume in notes or memory. + +### 6.8 Internal operations and developer tooling + +1. Fold `learning_ai_productivity_web` into a general-purpose internal operations shell or replace it with shared platform operator dashboards. +2. Expose `learning_ai_mac_tooling` findings into admin-web so security posture becomes visible next to auth, usage, and incident data. +3. Use local-llm-lab as the canonical place for model benchmarking before products adopt new local or hybrid model defaults. +4. Add a workspace-wide capability registry document generated from product manifests and package usage, not maintained manually. + +--- + +## 7. Improvement Ideas by Repo + +### `learning_ai_common_plat` + +- Add a cross-product artifact schema and artifact index service. +- Add a shared event taxonomy and event explorer dashboard. +- Add a trusted-device / device-risk service. +- Add a unified marketplace inventory model for templates, plugins, prompts, skills, agents, and workflows. +- Add a personal timeline API that aggregates cross-product activity. +- Add cross-product notification orchestration with channel rules and escalation rules. +- Add a reusable agent runtime contract covering session state, todos, plan state, resume, approvals, action logs, and handoff metadata. +- Add a bridge/dispatch abstraction for browser, IDE, mobile, and desktop task handoff. + +### `learning_voice_ai_agent` + +- Become the canonical capture service for voice-first input across the ecosystem. +- Export transcripts directly into NoteLett, MindLyst, and FlowMonk. +- Reuse ActionTrail to show transcript edits, exports, and agent transformations. + +### `learning_multimodal_memory_agents` + +- Become the canonical reflection and memory synthesis layer for notes, transcripts, habits, sessions, and routes. +- Accept imports from Cowork reports, PeakPulse sessions, and ChronoMind routines. +- Expose memory APIs consumable by JarvisJr and Local Memory GPT. + +### `learning_ai_clock` + +- Publish routines/reminder APIs that EffoRise, FlowMonk, JarvisJr, and NomGap can call. +- Share urgency and escalation models with auth approvals and agent approvals. +- Become the schedule engine for cross-product follow-through. + +### `learning_ai_fastgap` + +- Reuse EffoRise habits and ChronoMind routines rather than owning separate streak/reminder logic. +- Feed fasting state into coaching and reflection systems. +- Export milestone events into the shared personal timeline. + +### `learning_ai_jarvis_jr` + +- Reuse common platform marketplace instead of a product-local marketplace. +- Consume MindLyst memory and NoteLett notes as coaching context. +- Send delegated or automated actions through ActionTrail and Cowork-style policy gates. +- Reuse `claw-code` team/sub-agent, task, and persistent-memory patterns for coach specialization and delegated assistance. + +### `learning_ai_peakpulse` + +- Reuse MindLyst for post-activity reflection and journaling. +- Reuse ChronoMind for pacing alerts and preparation warnings. +- Feed route/session output into EffoRise effort scoring and shared activity timelines. + +### `learning_ai_notes` + +- Become the canonical editable workspace for artifacts generated by LysnrAI, Cowork, FlowMonk, and JarvisJr. +- Add first-class provenance and "created by agent" metadata from ActionTrail. +- Share block schemas with MindLyst and Local Memory GPT. +- Add browser and desktop quick-capture entrypoints modeled after Cowork's extension and dispatch patterns. + +### `learning_ai_flowmonk` + +- Build on ChronoMind routines, EffoRise habits, and NoteLett notes instead of reproducing those abstractions. +- Use Cowork for high-autonomy execution and ActionTrail for audit/approval. +- Export plans as artifacts and timeline events into common platform. +- Reuse `claw-code` todo lists, plan mode, compaction, and resume semantics so plans can survive long-running work without context loss. + +### `learning_ai_trails` + +- Become the ecosystem action ledger for every meaningful AI-driven operation. +- Add first-class adapters for Cowork, FlowMonk, JarvisJr, and LysnrAI transformations. +- Integrate device trust and auth state into approval policy decisions. +- Add replay views for browser capture, remote dispatch, scheduled runs, and resume chains across agent workflows. + +### `learning_ai_local_memory_gpt` + +- Serve as the privacy-preserving local cache and compare UI for selected ecosystem knowledge. +- Sync approved notes, transcripts, and summaries from cloud products. +- Reuse local-llm-lab model guidance and common platform artifact schemas. + +### `learning_ai_local_llms` + +- Become the standard evaluation and recommendation lab for local/hybrid model adoption across products. +- Publish a machine-readable capability matrix for models and hardware. +- Feed recommendation outputs into Local Memory GPT, Cowork, and any offline-capable products. + +### `learning_ai_smart_auth` + +- Move from spec-only posture toward a concrete reference architecture doc that maps directly to platform-service modules, auth-app features, and device trust integrations. + +### `learning_ai_auth_app` + +- Expand beyond login MFA into ecosystem action approval, budget escalation approval, and org admin confirmation. +- Add explicit support for approving agent runs and sensitive exports. + +### `learning_ai_efforise` + +- Reuse ChronoMind schedules, PeakPulse activity data, and MindLyst reflections to make habits feel contextual rather than isolated. + +### `learning_ai_productivity_web` + +- Either formalize as the lightweight internal ops sandbox or retire in favor of common platform dashboards and UX lab. +- If kept, use it as a thin shell over shared bridge, dispatch, and operator APIs rather than a separate island. + +### `learning_ai_mac_tooling` + +- Integrate with SmartAuth and platform-service as a device risk source. +- Feed exfiltration alerts into ActionTrail and admin-web. +- Use Cowork policies to limit high-risk local automation when the device is degraded. + +### `oss/learning_ai_claw-code-oss` + +- Reuse only the clean runtime/harness ideas that strengthen internal agent execution: session persistence, task/todo state, project memory, slash-command ergonomics, MCP lifecycle, config hierarchy, bridge modes, remote triggers, and worktree isolation. +- Do not allow the workspace to drift into a second ungoverned policy layer. + +### `oss/learning_ai_claw-cowork` + +- Integrate tightly with ActionTrail, SmartAuth, Auth App, and common platform marketplace. +- Emit structured artifacts into NoteLett and MindLyst after runs. +- Reuse local-llm-lab recommendations for local execution fallback and cost control. +- Promote its strongest reusable capabilities into the ecosystem: Dispatch API, scheduled automation, browser extension capture, audit panel patterns, prompt-injection detection, screenshot safety heuristics, and app/browser/desktop handoff. + +--- + +## 8. Priority Recommendations + +### P0: Foundational integrations + +1. Define a shared artifact schema in `learning_ai_common_plat`. +2. Define a shared event taxonomy and action log schema. +3. Define a unified approval model spanning SmartAuth, Auth App, ActionTrail, and Cowork. +4. Define a shared cross-product deep-linking convention. +5. Define a marketplace inventory model that supports plugins, agents, prompts, templates, and workflows. +6. Define a shared agent runtime contract for session memory, todo state, approvals, resume, dispatch, and schedule metadata. + +### P1: Cross-product user value + +1. LysnrAI transcript -> NoteLett note -> MindLyst memory pipeline. +2. FlowMonk plan -> ChronoMind routine -> EffoRise habit handoff. +3. Cowork run -> ActionTrail audit -> NoteLett report -> MindLyst memory ingestion. +4. PeakPulse session -> MindLyst reflection -> EffoRise effort credit. +5. Device risk from mac-tooling -> SmartAuth step-up auth -> Cowork execution restrictions. +6. Browser/web capture -> NoteLett or FlowMonk or MindLyst via a shared "Send to ..." entrypoint. +7. Remote task dispatch -> trusted desktop executor -> trails replay -> notes or memory output. + +### P2: Ecosystem differentiation + +1. Shared personal timeline across all products. +2. Shared notification and digest orchestration. +3. Shared creator marketplace across workflows, plugins, notes, routines, and coaching packs. +4. Hybrid cloud/local memory and model-routing layer combining common platform, Local LLM Lab, and Local Memory GPT. + +--- + +## 9. Execution Matrix + +| Initiative | Primary Owner | Key Dependent Repos | Impact | Effort | Why It Should Happen Early | +| ----------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ------ | ------ | ----------------------------------------------------------------------------------------------------- | +| Shared artifact schema | `learning_ai_common_plat` | `learning_voice_ai_agent`, `learning_ai_notes`, `learning_multimodal_memory_agents`, `learning_ai_flowmonk`, `oss/learning_ai_claw-cowork` | High | Medium | It unblocks memory, search, provenance, timelines, and agent output reuse | +| Shared event taxonomy + action log | `learning_ai_common_plat` + `learning_ai_trails` | `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, `learning_ai_jarvis_jr`, `learning_voice_ai_agent` | High | Medium | It turns fragmented agent/product telemetry into one reviewable substrate | +| Unified approvals and trust model | `learning_ai_smart_auth` + `learning_ai_auth_app` + `learning_ai_trails` | `oss/learning_ai_claw-cowork`, `learning_ai_mac_tooling`, `learning_ai_common_plat` | High | Medium | It creates one policy model for login, step-up auth, and agent approvals | +| Shared agent runtime contract | `learning_ai_common_plat` | `oss/learning_ai_claw-code-oss`, `oss/learning_ai_claw-cowork`, `learning_ai_flowmonk`, `learning_ai_jarvis_jr`, `learning_ai_trails` | High | Medium | It prevents each agent product from reinventing session state, todos, approvals, and resume flows | +| Transcript -> note -> memory pipeline | `learning_voice_ai_agent` + `learning_ai_notes` + `learning_multimodal_memory_agents` | `learning_ai_common_plat` | High | Medium | It creates immediate user-visible value across three flagship products | +| Plan -> routine -> habit handoff | `learning_ai_flowmonk` + `learning_ai_clock` + `learning_ai_efforise` | `learning_ai_common_plat` | High | Medium | It turns planning into follow-through instead of isolated planning data | +| Cowork -> Trail -> Notes -> Memory flow | `oss/learning_ai_claw-cowork` + `learning_ai_trails` + `learning_ai_notes` + `learning_multimodal_memory_agents` | `learning_ai_common_plat` | High | High | It creates a differentiated audited-agent workflow that few ecosystems have | +| Device trust ingestion | `learning_ai_mac_tooling` + `learning_ai_common_plat` | `learning_ai_smart_auth`, `learning_ai_auth_app`, `oss/learning_ai_claw-cowork` | Medium | Medium | It upgrades desktop/agent safety with existing assets already in the workspace | +| Shared browser and desktop dispatch layer | `learning_ai_common_plat` + `oss/learning_ai_claw-cowork` | `learning_ai_notes`, `learning_ai_flowmonk`, `learning_multimodal_memory_agents`, `learning_ai_trails` | High | Medium | It lowers capture friction and creates one route from browser or mobile into agent-assisted workflows | +| Shared marketplace inventory model | `learning_ai_common_plat` | `learning_ai_jarvis_jr`, `oss/learning_ai_claw-cowork`, `learning_ai_notes`, `learning_ai_flowmonk`, `learning_ai_clock` | High | High | It can unify monetization and reusable asset distribution | +| Shared personal timeline | `learning_ai_common_plat` | Almost all product repos | High | High | It becomes the ecosystem shell once events and artifacts are standardized | +| Hybrid local/cloud AI routing | `learning_ai_local_llms` + `learning_ai_local_memory_gpt` + `learning_ai_common_plat` | `oss/learning_ai_claw-cowork`, `learning_multimodal_memory_agents`, `learning_voice_ai_agent` | Medium | High | It is strategically valuable, but depends on artifact and policy standardization first | + +--- + +## 10. Guardrails: What Not to Over-Unify + +Cross-pollination only helps if it centralizes the right seams. These areas should stay product-specific: + +1. Product voice, branding, and UX metaphors should remain local to each repo. +2. Domain logic should remain local when it is the core product moat. Examples: fasting logic in `learning_ai_fastgap`, adventure session semantics in `learning_ai_peakpulse`, timer cascades in `learning_ai_clock`. +3. Local-first privacy behavior in `learning_ai_local_memory_gpt` should not be weakened just to mirror cloud-first products. +4. Agent runtime experimentation in `oss/learning_ai_claw-code-oss` and `oss/learning_ai_claw-cowork` should stay decoupled from platform policy enough to evolve safely. +5. Shared platform should own contracts, policy, schemas, and orchestration primitives, but not erase the reason each product exists. + +The unification target should be: + +- identity +- approvals +- events +- artifacts +- runtime contracts +- dispatch and bridge contracts +- notifications +- marketplace inventory +- retrieval contracts +- observability + +The non-target should be each app's core domain experience. + +--- + +## 11. Suggested 90-Day Sequence + +### Days 1-30 + +1. Draft the shared artifact schema. +2. Draft the shared event taxonomy and action log schema. +3. Draft the unified approval and trusted-device model. +4. Draft the cross-product deep-link convention. +5. Draft the shared agent runtime contract for session state, resume, todos, schedules, and dispatch. + +### Days 31-60 + +1. Implement the LysnrAI -> NoteLett -> MindLyst pipeline. +2. Implement the FlowMonk -> ChronoMind -> EffoRise handoff. +3. Implement the Cowork -> ActionTrail audit event emission path. +4. Prototype mac-tooling -> platform-service device risk ingestion. +5. Prototype shared browser capture and dispatch using Cowork extension and Dispatch API patterns. + +### Days 61-90 + +1. Stand up a personal timeline API and internal timeline viewer. +2. Stand up a shared notification orchestration service. +3. Convert marketplace concepts into one inventory model and one entitlement model. +4. Add one operator cockpit view that merges auth, usage, risk, and AI action streams. +5. Add resume-anywhere flows across desktop, auth app, trails, and notes or memory consumers. + +--- + +## 12. Recommended Next Documents + +This document should lead to a small set of follow-on specs rather than another broad audit: + +1. `docs/ECOSYSTEM_SHARED_ARTIFACT_SCHEMA.md` +2. `docs/ECOSYSTEM_EVENT_TAXONOMY.md` +3. `docs/ECOSYSTEM_APPROVALS_AND_TRUST_MODEL.md` +4. `docs/ECOSYSTEM_MARKETPLACE_UNIFICATION.md` +5. `docs/ECOSYSTEM_PERSONAL_TIMELINE_PRD.md` + +--- + +## 13. Bottom Line + +The ecosystem's biggest unrealized strength is not any single app. It is the combination of: + +- strong identity and approvals +- strong voice and multimodal capture +- strong memory and note systems +- strong time/habit/planning loops +- strong agent execution and audit infrastructure +- strong harness and runtime patterns already proven in `claw-code` and `claw-cowork` + +The main gap is that these strengths still live in adjacent repos instead of one intentional platform model. The next step should be standardizing the shared contracts that let each repo keep its own product identity while participating in one ByteLyst operating system. diff --git a/docs/ecosystem/README.md b/docs/ecosystem/README.md new file mode 100644 index 00000000..7a859889 --- /dev/null +++ b/docs/ecosystem/README.md @@ -0,0 +1,28 @@ +# Ecosystem Docs + +Canonical home for ByteLyst ecosystem-wide architecture and rollout documentation. + +Use this folder for: + +- ecosystem strategy +- shared contracts and schemas +- cross-repo ownership and adoption plans +- rollout sequencing +- verification matrices + +Recommended structure: + +- [`ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md`](/Users/saravana/BytelystAI/learning_ai/learning_ai_common_plat/docs/ecosystem/ECOSYSTEM_CROSS_POLLINATION_OPPORTUNITIES.md) — strategy, opportunities, gaps, priorities +- `ECOSYSTEM_SHARED_ARTIFACT_SCHEMA.md` — canonical artifact model +- `ECOSYSTEM_EVENT_TAXONOMY.md` — shared events and action-log contract +- `ECOSYSTEM_AGENT_RUNTIME_CONTRACT.md` — shared session, todo, resume, dispatch semantics +- `ECOSYSTEM_APPROVALS_AND_TRUST_MODEL.md` — approvals, trust levels, device posture +- `ECOSYSTEM_MARKETPLACE_UNIFICATION.md` — shared marketplace model +- `ECOSYSTEM_PERSONAL_TIMELINE_PRD.md` — unified user activity stream +- [`adoption/`](/Users/saravana/BytelystAI/learning_ai/learning_ai_common_plat/docs/ecosystem/adoption) — per-product adoption notes + +Ownership model: + +- `learning_ai_common_plat` holds the canonical ecosystem docs +- product repos hold repo-local adoption and implementation details +- `claw-code-oss` and `claw-cowork` remain reference sources for runtime and execution patterns diff --git a/docs/ecosystem/adoption/README.md b/docs/ecosystem/adoption/README.md new file mode 100644 index 00000000..61de5c87 --- /dev/null +++ b/docs/ecosystem/adoption/README.md @@ -0,0 +1,21 @@ +# Adoption Docs + +Use this folder for per-product ecosystem adoption notes. + +Recommended file pattern: + +- `learning_voice_ai_agent.md` +- `learning_multimodal_memory_agents.md` +- `learning_ai_clock.md` +- `learning_ai_notes.md` +- `learning_ai_flowmonk.md` +- `learning_ai_trails.md` +- `oss_learning_ai_claw_cowork.md` + +Each adoption doc should cover: + +- shared contracts consumed +- local UX changes +- migration plan +- open risks +- verification checklist