Roadmap

What's shipped, what's building, what's next.

The runtime is real and running in production today. The self-serve deploy loop is what we're building toward general availability. We build in the open, so here's the whole board.

Shipped live today

Drop-in agent API

Speaks the LangGraph Server protocol, so langgraph_sdk, RemoteGraph, and Studio connect unchanged.

Durable state

Postgres checkpoints. Threads survive restarts and resume mid-run.

Crash-safe run queue

Background runs with a visibility deadline; orphaned runs are reclaimed and resumed.

Streaming

values, updates, messages, and full event streams over SSE.

Human-in-the-loop

Dynamic interrupt, inspect, and resume with a command.

Time travel, webhooks, store, cron

Checkpoint history and run-from-checkpoint, run-outcome webhooks, long-term cross-thread memory, scheduled runs.

Metering & auth

Per-account usage ledger and x-api-key auth, now with per-run active-CPU, wall, and step instrumentation.

Building toward GA

Multi-tenant runtime

One runtime hosting many deployments, with state, usage, and checkpoints scoped per account. The critical path.

Accounts, projects, and API keys from a GitHub login.

Push to deploy

A repo with a buildstage.json builds and deploys to a live URL.

Deploy GitHub Action

buildstage/deploy@v1 deploys on every push. No CLI.

Control-plane dashboard

Deployments, usage, logs, and run history in one place.

Quota & isolation

Free-tier caps and a tenant-isolation pass: no account can read another's data.

Planned after GA

Self-serve billing

Rating and invoicing on top of the usage ledger.

Teams & roles

Shared projects, members, and permissions.

Rollbacks

Instant revert to a previous deployment.

More frameworks

Adapters for CrewAI, AutoGen, Google ADK, and the Claude Agent SDK. Any graph deploys the same way.

AWS migration

Move the worker to containers first, keeping the API portable, as volume justifies it.

Committed & self-host

Volume pricing and run-in-your-own-cloud for teams that need it.