Post-alpha roadmap

Alpha: @statelyai/agent 2.0 is in alpha. APIs can change between releases; pin an exact version. Feedback: github.com/statelyai/agent.

What gates 2.0 stable, and what is honestly not shipped yet. Everything not listed as gating is additive and waits on real usage feedback; if one of these blocks you, open an issue.

Road to v2 stable (gating)

v2 leaves alpha only when its durable formats can be promised. Changing a persisted format after stable is worse than staying alpha longer, so these land in order and formats freeze last.

• Runtime semantics complete first. The durable-execution seams still moving (step request envelope, execution info, event-log contract, dynamic actor binding) settle before any freeze.
• Stable XState v6. The alpha pins xstate@6.0.0-alpha.*. The persisted snapshot format is XState's, so a stable v6 is a hard dependency: if it slips, v2 stays alpha rather than freezing on an alpha format.
• Node LTS matrix. Declare and CI-test the supported active LTS versions.
• Format freeze. The durable formats freeze together, with documented versioning rules: the AgentLogEntry envelope (AGENT_EVENT_SCHEMA_VERSION, including the reserved @agent.init event and verification hashes), the AgentEventLogStore protocol, and the AgentTraceEvent envelope (AGENT_TRACE_SCHEMA_VERSION). Persisted snapshots are compaction caches, versioned by XState.
• Upgrade tests. Fixture snapshots and traces from the frozen formats, replayed in CI against every release, so a break is caught before it ships.
• RC cycle. At least one release candidate with formats frozen and real hosts running against it before 2.0.0.

Not shipped yet

• Postgres and Redis storage adapters. Core ships the persistence contracts, an in-memory event-log store, and SQLite (createSqliteEventLogStore / createSqliteSnapshotStore from @statelyai/agent/sqlite). No Postgres or Redis packages.
• OpenTelemetry exporter. @statelyai/agent/otel ships createOtelTraceHandler (trace-to-span mapping), but no exporter or SDK lifecycle; bring your own provider, or build on onTrace with serializeTraceEvent(event) for JSONL-safe output.
• SSE/WebSocket transport helpers. Host your own stream over onChunk (an SSE example ships in examples/sse-transport).
• Agent-specific dynamic fan-out helper. Fan-out works today via XState spawn(...) or Promise.all(...); core has no higher-level helper for branch binding and progress.
• Visualization tooling. Stately Studio and the VS Code extension own diagramming and inspection.

Near-term, non-gating

• Managed step-path helper. A collapsed driver over the step path loop. Deferred because the loop is ~15 lines of host-owned code by design.
• Idle persist/revive helper. The persist, return handle, resume-with-event recipe (see human in the loop) is a documented pattern each host rewrites; a helper lands once real stores show the common shape.
• Plan executor. Multi-event commands are an explicit decide loop today; the executor layer that simulates a proposed plan up front and replans on divergence is next, as a documented pattern first.
• Live-path mid-flight resume for fan-out. Event-log replay already re-derives still-owed spawned effects; restoring a live runAgent snapshot persisted mid-flight still drops frozen children.