The Roadmap

The core infrastructure is already in place and operational.

• Engram (knowledge-service/) — Rust gRPC, 139 tests, ✅ COMPLETE. THE MVP CORE. Everything builds from here.
• Exodia — VPS daemon, Agent Mail MCP, 120+ commits, ✅ COMPLETE.
• OMO fork — 45 custom commits: Coeus planning, hooks, taxonomy, 46 lifecycle hooks, ✅ COMPLETE.
• beads/bd/bv — issue tracking + graph-aware triage, ✅ COMPLETE.

graph TB
    subgraph DONE["✅ DONE / EXISTS"]
        K1["knowledge-service/
Rust gRPC · 139 tests"]
        K2["knowledge-service-mcp/
6 MCP tools · Rust"]
        K3["oh-my-opencode/
41 hooks · 7 events"]
        K4[".beads/
Git-native issue tracking"]
        K5["exodia/
Python VPS daemon"]
    end

    subgraph REF["📚 REFERENCE (use as-is, don't modify)"]
        R1["mastra/packages/core/
Workflow engine"]
        R2["mastra/MCPClient
Multi-server aggregation"]
        R3["goose/
Rust binary, HTTP SSE"]
        R4["automaker/
Port: GitDiffPanel, CreatePRDialog"]
    end

    subgraph NEW["🔧 NEW (must build)"]
        N1["packages/blueprint-engine/
DSL + nodes + library + API"]
        N2["toolshed/
Config + profiles"]
        N3["toolshed/servers/bd-mcp/"]
        N4["toolshed/servers/git-mcp/"]
        N5["toolshed/servers/quality-mcp/"]
        N6["toolshed/servers/context-mcp/"]
        N7["packages/cli/
jarvis command"]
    end

    subgraph EVOLVE["🔄 EVOLVE (existing, add features)"]
        E1["autoboard/apps/web
+ split-panel review
+ diff viewer
+ context panel
+ blueprint progress"]
        E2["autoboard/apps/api
+ PR creation
+ webhook events
+ context snapshots"]
        E3["autoboard/packages/git/
Port from automaker"]
    end
        

• Goal: Maximize OMO + CLI subscription tokens.
• Critical path: knowledge-service-mcp → OMO hooks → knowledge flows.
• The bottleneck: Wire Engram via knowledge-service-mcp (1,014-line plan ready to build).
• Nightshift: overnight autonomous runs for Phase 1 heavy lifting.
• Economics: Claude Pro $20/mo + ChatGPT $20/mo subscriptions — cheaper for single-user heavy dev work.
• What gets wired: knowledge-service-mcp (MCP server wrapping Engram), OMO hooks for auto-recall/capture.

Phase 2 — Blueprint Engine

• Goal: Deterministic + agent node interleaving.
• Blueprint Engine: DSL (~500 LOC TypeScript) + Mastra workflow + Goose agent harness.
• Mastra (workflow engine): studied as reference implementation, wired in this phase.
• Goose (agent harness): studied as reference implementation, wired in this phase.
• Autoboard: split-panel review UI for blueprint outputs.
• jarvis CLI: command-line interface for blueprint submission.

Phase 3 — Org-Level Platform

• Goal: Multi-tenant, multi-model, org-scale.
• OpenRouter: when multi-tenant justifies per-token billing + 5.5% fee.
• Multi-tenant auth + RBAC
• Lamport-ordered concurrency, Paxos merge consensus
• Slack/Discord: via OpenClaw patterns (studied in Phase 1, wired in Phase 3).
• Governance: via Paperclip patterns (studied in Phase 1, wired in Phase 3).

Studied But NOT Integrated

Reference implementations mapped to their phases. These are studied for patterns, wired only when their phase arrives.

• TigerBeetle → quality standard for ALL phases (not a deployment target).
• OpenClaw → chat interface patterns for Phase 3.
• Paperclip → governance patterns for Phase 3.
• Nightshift → used as-is for Phase 1 overnight, studied for task taxonomy patterns.

Deferred Decisions

Open questions explicitly deferred for future consideration (to be addressed later):

These future items are not part of the current roadmap.

• Multi-user auth (when does multi-user matter?)
• Horizontal scaling (managed PostgreSQL, Qdrant as dedicated vector search)
• Embedding provider choice (OpenAI text-embedding-3-small vs local Ollama)
• Productization: billing, docs, support
• Canonical knowledge seeding (CS algorithms taxonomy? Design patterns?)