docs/CODEMAP.md
By Alex Place · Updated 2026-06-20

Keystone OS — Codemap & Feature Roadmap

Canonical status reference for the Keystone OS codebase. Version: 2026-06-07 Coverage: System-wide feature state, active surfaces, and forward roadmap.


How to Read This Document

  • Implemented — Code exists, tests pass, CI green, converges clean.
  • Active — In development, has open branch or TODO, may have known gaps.
  • Queued — Spec'd, slotted in agent backlog, not yet started.
  • Held — Blocked on dependency, external system, or operator decision.
  • Retired — Removed from codebase or superseded.

1. System Overview

Keystone OS is a local-first operating cockpit combining:

Surface Technology Purpose
Lantern Garage Node.js, vanilla JS, HTML Web app, API, PWA, static assets
Dream Journal Browser (localStorage + JSONL export) Conversational dream capture and chat
Agent Fleet PowerShell + Node.js convergence loop Validation, inspection, batch processing
RAG House Markdown flat files + JSON manifests Evidence-backed retrieval and research
CSF Memory Python (src/csf/) + JSON schemas Symbolic memory, compression, provenance
Convergence IO PowerShell + Node.js 12-step validation loop with receipts
Discord Bot Node.js (optional) Community tier integration

Operating model:


capture context

  -> classify work (DCF)

  -> route through convergence (PCSF)

  -> validate with receipts (AAPF)

  -> store evidence in RAG / CSF

  -> promote, hold, or archive


2. Feature Inventory

2.1 Lantern Garage Web Server

Feature Status File Notes
HTTP server Implemented apps/lantern-garage/server.js Port configurable via LANTERN_GARAGE_PORT
PWA manifest Implemented apps/lantern-garage/public/manifest.json Installable surface
Health endpoint Implemented /api/health Returns { ok, service, generatedAt }
Status endpoint Implemented /api/status Full system state (arc, wallet, controls, readiness)
Readiness gates Implemented /api/readiness Dual-boot prep status
Action capabilities Implemented /api/action-capabilities Available actions with real/held classification
Chat (non-streaming) Implemented /api/dream/chat POST; Gemini grounding, Anthropic, OpenAI, Ollama, orerror
Chat (streaming SSE) Implemented /api/dream/chat/stream Server-sent events with provider fallback
Dreamer CRUD Implemented /api/dreamer Create, read dream entries
Dreamer chat Implemented /api/dreamer/chat Dream + chat combined endpoint
Agent list Implemented /api/agents Returns persona roster
Agent slots Implemented /api/agents/slots Reads from manifests/dream-journal-v1-agent-slots.json
Keystone status Implemented /api/keystone/status Git branch, dirty files, provider key presence
Agent Status page Implemented public/agent-status.html Dashboard linking to all status endpoints
Knowledge Center Implemented public/knowledgecenter.html Documentation index
Theme toggle Implemented public/index.html Dark/light mode
Offline chat fallback Retired lib/dream-chat.js Removed June— now returnserror with help text

2.2 Dream Journal

Feature Status Notes
Freeform chat journaling Implemented Text input, multi-turn, local storage
Persona routing Implemented 6 personas: Keystone, Lantern, Xenon, Blinkbug, Waterfall, Founder
Door parsing Implemented `[DOORS: A B C]` parsed and rendered as chips
Voice input (Web Speech API) Implemented Browser TTS; voice selector in settings
Symbol tagging Implemented DCF auto-classification on entries
JSONL export Implemented Portable local export for CSF ingestion
3 Door Game Design Implemented Spec complete; playable frontend pending
STT backend fallback Queued Vosk route POST /api/dream/transcribe planned
CTF glyph persistence Implemented Symbols stored in entries, surfaced in stats
TTS voice selector Implemented Rate/pitch/name from SpeechSynthesis.getVoices

2.3 Convergence & Validation

Feature Status Notes
12-step convergence loop Implemented PowerShell: scripts/Invoke-LanternSmartConvergenceLoop.ps1
Git state checks Implemented Dirty working tree, locks, divergence
npm audit Implemented Runs in Standard+ depth; skipped in Light
Health probe Implemented Local HTTP probe every run
Self-repair (ApplySafeFixes) Implemented Manual trigger only after clean git scan
Validation ring Active 3-agent consensus, SHA-256 hash chain, tamper-evident JSONL
Fleet status reporting Implemented Reads data/status/super-jarvis-fleet.json
Task intake queue Implemented mcp1_task_intake / mcp1_queue_status
Convergence receipts Implemented JSON output with status, issue count, next action

2.4 Provider & AI Gateway

Feature Status Notes
Gemini API with grounding Implemented google_search_retrieval tool in payload
Anthropic Claude Implemented api.anthropic.com messages endpoint
OpenAI GPT Implemented api.openai.com chat completions
Ollama local Implemented http://127.0.0.1:11434 fallback
PCSF live refresh Implemented Auto-updates data/pcsf/*.pcsf.json on server start
Provider state caching Implemented 60s PCSF cache in lib/provider-cache.js
Multi-provider fallback chain Implemented Gemini -> Claude -> OpenAI -> Ollama -> 503

2.5 RAG & Research

Feature Status Notes
Flat RAG dollhouse Implemented skills/lantern-rag-dollhouse/
External LLM web cache Implemented data/rag-intake/external-llm-web-cache/
Internal RAG house Implemented data/internal-rag-house/
Source registry pattern Implemented From archived gm-agent-orchestrator; active in research workflow
Claim registry pattern Implemented JSONL machine-readable claim ledger
Research synthesis Active First synthesis completed: human-in-the-loop ASI/AGI
RAG intake queue Active data/rag-intake/ with trace and downloads-merge

2.6 CSF Memory Engine

Feature Status Notes
Core memory tiers Implemented EPHEMERAL, WORKING, REFINED, SKILL, ARCHIVE
Procedural tier Implemented Added June 2026: create_procedure(), steps, tool_invocations
Multi-signal retrieval Implemented Semantic + BM25 keyword + entity overlap fusion
Async writes Implemented write_async() with queue depth tracking
Actor-aware provenance Implemented actor_id, actor_type, confidence_reasoning
Staleness detection Implemented Contradiction check against same-entity memories
Temporal lineage Implemented promotion_lineage with evolution events
Integrity hashes Implemented SHA-256 on AAPF records

3. Forward Roadmap

Phase 1: v1.0.0 — Foundation Complete (June 2026)

Goal: Solid core with provider gateway, convergence validation, and local-first architecture.

  • Lantern Garage server with full API surface
  • Dream Journal chat with multi-persona routing
  • Convergence loop with 12-step validation
  • PCSF/CCF/NAP/DCF/AAPF regulatory primitives
  • Agent Status dashboard
  • Knowledge Center documentation index
  • Multi-provider AI gateway (Gemini, Claude, OpenAI, Ollama)
  • Gemini grounding with Google Search retrieval
  • CSF memory engine v2 (async, multi-signal, procedural tier)
  • RAG dollhouse and external web cache

Remaining before tag:

  • Configure real provider API keys in .env
  • Verify Gemini grounding with paid key (dream_journal/gemini_grounding_test)
  • Final convergence pass with all green
  • Tag v1.0.0 and create release notes

Phase 2: v1.0.1 — Interactive Depth (Late June / Early July 2026)

Goal: Make the journal alive. Add voice, interactive gameplay, and deeper symbolic tools.

  • FullDoor Game with Information Cards (frontend + backend)
  • Backend Vosk STT fallback (POST /api/dream/transcribe)
  • Enhanced Discord bot integration with tier-aware roles
  • Advanced symbolic tools (void/light balance, mystery tracking dashboard)
  • Dream pattern analysis dashboard (web UI)
  • Mobile-friendly responsive UI pass
  • Full DCF privacy controls + user-facing export tools

Phase 3: v1.0.2 — Research & Intelligence (July 2026)

Goal: Turn Keystone OS into a research cockpit, not just a journal.

  • Research request pipeline: research/requests/open/ with structured templates
  • Automated source registry freshness checks
  • Claim registry auto-update from synthesis documents
  • Web search integration for real-time research intake
  • Research synthesis generation from web cache entries
  • Convergence loop refinements: backpressure, rate limiting, OpenTelemetry spans
  • Lucid dreaming tools (MILD/WBTB tracking, reality check reminders)

Phase 4: v1.1 — Guild & Team Features (August 2026)

Goal: Team collaboration, shared archives, and advanced audit.

  • Team / multi-user support for Synthesasia Guild tier
  • Shared symbolic archives with access controls
  • Blockchain-anchored AAPF audit trails (optional for Guild)
  • Custom art style training pipeline
  • API access for Guild members (documented endpoints)
  • Advanced analytics and reporting (skills/lantern-custom-report-lib/)

Phase 5: v1.2 — Convergence Intelligence (September 2026)

Goal: The system improves itself through co-improvement with the operator.

  • Agent swarm orchestration with human-in-the-loop steering
  • Self-monitoring convergence metrics (drift detection, regression alerts)
  • Automated research intake from configured RSS feeds and arXiv alerts
  • Skill crystallization: CSF procedural memories auto-promote to reusable skills
  • Co-improvement loop: agent proposes → operator judges → agent implements → convergence validates

4. Active Workstreams

Workstream Lead Agent Status Next Action
Chat provider gateway Xenon Implemented Configure real API keys
Agent Status dashboard Keystone Implemented Add real-time websocket updates
Convergence validation ring Founder's Wish Active Stabilize 3-agent consensus rate
CSF memory engine v2 Blinkbug Implemented Add test coverage for create_procedure()
RAG research synthesis Waterfall Active Fill inaccessible paywalled sources
3 Door Game Gage Queued Frontend implementation
Discord integration Xenon Queued Role + channel setup with tier enforcement
Cloud deployment Tony Held Waiting on dual-boot readiness

5. Dependencies & Blockers

Dependency Blocks Resolution Path
Real Gemini API key gemini_grounding_test Operator adds GEMINI_API_KEY to .env
Real Anthropic key Claude fallback reliability Operator adds ANTHROPIC_API_KEY to .env
Ollama installed locally Offline AI capability Run scripts/orchestration/start-local-llms.ps1
Dual-boot readiness Cloud deployment Track in manifests/validation/DUAL-BOOT-PREP-LATEST.json
GitHub Actions secrets CI deployment Operator configures DISCORD_TOKEN, deploy keys

Document Path What It Covers
Architecture (current state) docs/ARCHITECTURE.md Canonical "what the system is today" snapshot
Architecture Decision Records docs/adr/README.md Decision log — why the architecture is the way it is
Dream Journal Roadmap docs/DREAM-JOURNAL-ROADMAP.md Dream Journal specific features, tiers,Door Game
Research Contract archive/gm-agent-orchestrator/docs/research/research-contract.md How research is ingested, audited, and promoted
Repo Contract docs/REPO-CONTRACT.md Git workflow, convergence rules, merge policy
Convergence Loop docs/CONVERGENCE-CICD-BATCH-LOOP.md 12-step method, CI integration, batch framework
Agent Fleet Plan manifests/CONVERGENCE-LOOP-AGENT-FLEET.md 36-slot matrix, 64-worker elastic target
RAG Dollhouse Skill skills/lantern-rag-dollhouse/SKILL.md Flat RAG building, asset manifest, purge lanes
Latest Research Synthesis data/rag-intake/research-synthesis-human-in-the-loop-asi-agi-2026-06-07.md Human-in-the-loop ASI/AGI self-improvement

7. Maintenance Rules

  1. Update this Codemap whenever a workstream status changes (implemented, active, queued, held, retired).
  2. Do not duplicate the Dream Journal Roadmap — link to it for tier-specific and game-specific features.
  3. Add new features to Sectionbefore promoting them to the roadmap in Section 3.
  4. Retire features by moving them to a "Retired" subsection with a superseded-by note.
  5. Commit on change — this is a canonical doc, not a scratchpad.

This Codemap is the single source of truth for Keystone OS feature state. When in doubt, check here first.