R1 Roadmap

Product Stack

R1 (open source)        AI coding orchestrator. Works everywhere.
Ember (SaaS)            Cloud dev machines. Where R1 runs best.
Flare (infrastructure)  Firecracker microVMs on GCP. Powers both.

Current State (v2)

R1 is a Go binary with two layers:

V1 Execution Engine — Wraps Claude Code and Codex CLI as execution engines.

Deterministic PLAN -> EXECUTE -> VERIFY -> COMMIT phases with multi-model routing,

parallel agent coordination via git worktrees, and structured quality gates.

V2 Governance Layer — Multi-role consensus architecture:

Append-only content-addressed ledger (immutable reasoning graph)
Durable WAL-backed event bus with hooks and causality tracking
30 deterministic supervisor rules across 10 categories
7-state consensus loops for structured agreement
10 stance roles (PO, CTO, QA Lead, etc.) with per-role tool authorization
Bridge adapters wiring v1 systems into v2 event bus + ledger

MIT license

Ember Integration

Phase 1: Config + Managed AI — Implemented

internal/env/ember/ember.go (458 LOC) + internal/env/ember/ai.go (197 LOC)

Routes through Ember /v1/ai/chat API via internal/provider/ember.go (161 LOC)
OpenRouter base rate + 20% markup, billed to Ember account
Model routing: user subs first, then user API keys, then managed pool

Configuration in ~/.stoke/config.yaml:

ember:
  key: "ek_live_..."
  managed_models:
    enabled: true

Phase 2: Burst Compute — Implemented

internal/env/ember/ember.go includes burst compute integration.

Spawns Flare VMs via Ember /v1/workers API
Workers appear on Ember dashboard grouped under parent machine
Auto-destroy on task completion
Scheduler decides local vs burst based on estimated task duration

Configuration:

ember:
  burst:
    enabled: true
    threshold_minutes: 5
    worker_size: "4x"
    max_workers: 8
    auto_destroy: true

Phase 3: Remote Progress — Implemented

internal/remote/session.go (170 LOC)

Pushes progress to Ember /v1/sessions API
Live progress at ember.dev/s/<session_id>
Shareable URL for team visibility

Phase 4: Chat Sidebar — Not in scope for R1

The Ember desktop app / VS Code extension wraps R1's plan generation

in a conversational interface. This is an Ember-side workstream. The R1

side already exports the surfaces it needs via the MCP server and HTTP API.

Flare Integration

R1 talks to the Fly-compatible REST API via internal/env/fly/ which

works against Flare unchanged. No additional R1 work is needed for

Flare compute backends.

Open Source Strategy

License: MIT
R1 works fully standalone (no Ember required)
Compute backend interface is extensible (inproc, docker, ssh, fly, ember)
The value is the Ember API behind the key, not the client code

R1 Roadmap

R1 Roadmap

Product Stack

Current State (v2)

Ember Integration

Phase 1: Config + Managed AI — Implemented

Phase 2: Burst Compute — Implemented

Phase 3: Remote Progress — Implemented

Phase 4: Chat Sidebar — Not in scope for R1

Flare Integration

Open Source Strategy

Pages in this directory