Skip to content

Latest commit

 

History

History
83 lines (65 loc) · 3.42 KB

File metadata and controls

83 lines (65 loc) · 3.42 KB

Roadmaps

Per-subsystem delivery plans. Checkbox documents, not design docs.

For each subsystem we have:

  • A design doc in docs/<SYSTEM>.md — the why and what.
  • A roadmap in docs/roadmaps/<SYSTEM>.md — the when and the what's next.

Start with the design doc. Come here to see what's done, what's in progress, and what's next.

Index

Roadmap Subsystem Design doc
ENGINE.md Core engine ../ENGINE.md
STORES.md Stores (Org→Store→Docs) ../STORES.md
PROFILES.md Domain-aware structuring ../PROFILES.md
SERVER.md HTTP/gRPC transport ../SERVER.md
LLMGATE.md LLM gateway library ../LLMGATE.md
CONTROL-PLANE.md SaaS backend ../CONTROL-PLANE.md
DASHBOARD.md Web UI ../DASHBOARD.md
SDKS.md TS / Python / Go clients ../SDKS.md
MCP.md MCP adapter ../MCP.md
DEPLOYMENT.md Infra + CI/CD ../DEPLOYMENT.md

How these relate to the root ROADMAP.md

The file at the repo root (/ROADMAP.md) is the engine's working roadmap — the checkbox list used during active engine development. It's a superset of ENGINE.md for now because the engine repo is the only one that exists yet.

When repos split out (vectorless-server, llmgate, etc.), each will grow its own ROADMAP.md at its root, and the corresponding file here becomes a pointer.

Legend

All roadmaps use the same symbols so a skim tells you status instantly:

  • [x] — done, shipped.
  • [~] — in progress, actively being worked.
  • [ ] — not started, committed.
  • [?] — idea, plausible but not committed.
  • (opt) — optional polish, nice-to-have.

Conventions

  • Before starting a task: flip [ ] -> [~] in a tiny commit so collaborators see it's claimed.
  • On merge: flip [~] -> [x] in the same PR that delivers the work.
  • New ideas: drop them in with [?] under the right phase. No defensive TODOs — if you wouldn't bet a week on it being built, leave it as [?] rather than [ ].
  • Removals: if a task turns out not to make sense, delete it rather than leaving a zombie checkbox. Git history keeps the decision record.
  • No dates. Phases are sequenced, not scheduled. Dates are a lie. If external commitments demand a timeline, track those in an issue, not here.

Phase numbering

Each subsystem has its own phase numbering (its own "Phase 0" scaffold phase, its own "Phase 1", etc.). A phase in one subsystem does not correspond to the same phase in another — e.g. the control plane's Phase 0 happens long after the engine's Phase 2.

The rough ordering of cross-subsystem work, at the project level:

  1. Engine phases 0–2 (scaffold, ingest, retrieval) — where we are now.
  2. Extract llmgate into its own repo.
  3. Extract vectorless-server into its own repo.
  4. SDKs phase 0 (generate from proto).
  5. MCP phase 0 (tiny adapter over the TS SDK).
  6. Deployment phase 0 (one-region SaaS on Fly + Neon + R2).
  7. Control plane phase 0 + Dashboard phase 0 (auth, keys, usage).
  8. Billing, plans, Stripe — control plane phase 1.
  9. Engine phase 3+ (ecosystem, scale).