Skip to main content
Durable runtime for AI agents

Durable orchestration for agentic workflows.

The topology you build changes every quarter. The runtime underneath doesn’t. Smithers persists every step, so a crash is a resume point, not lost work, and a long job is something you walk away from. Any model, any harness, open source.

Read the docsDocs for agents →
Run this in your project
$bunx smithers-orchestrator init
Creates an editable hello.mdx workflow you can run immediately, then installs the smithers skill into your agents.
ship-it · a7b93f2
running
you > Research, plan, then implement. Pause for my approval before you ship.
✓researchpersisted
✓planpersisted
↻implement (resumed)running
⏸approval to shipwaiting
crash-safe · resumes from the last persisted step
Model and harness agnostic

Point each task at whichever agent and model is best for the job. Swap the harness without rewriting the workflow.

Why Smithers

The right way to build an agent
changes every six months.

Chains. ReAct. Crews. Swarms. Background agents. Subagent fan-out. Couple your infrastructure to any one of them and you’ve already rebuilt twice. One layer underneath never changes: durable orchestration. Smithers is that layer.

Claude, GPT, Gemini, Kimi
Volatile · weekly
ReAct, crew, swarm, background agents
Fluid · quarterly
Durable steps, retries, state, approvals, observability
Stable · doesn’t change
Every step is checkpointed the moment it finishes. A crash, a closed laptop, a flaky tool: all resume points.
Review loops, panels, debates, sagas. Named patterns are small TSX components on the substrate. Read them, fork them.
Every transition, attempt, and approval is a SQLite row. Prometheus metrics and traces with no setup.

Two ways in.

For humans

Run one command and ask the agent you already use for an outcome. Learn what Smithers is, then drive it without writing a workflow by hand.

Read the guide →
For agents

The full API, components, and patterns, written for the coding agent that authors and runs your workflows. Or feed it /llms-full.txt directly.

Read the agent docs →

MIT-licensed and open source on GitHub.