Frequently asked
What evaluators ask before they adopt.
No marketing answers. If a question evaluators raise has a different answer than the one below, the docs at docs.division.sh are binding, the platform specification is binding above that.
- How is this different from other agent frameworks?
- Most agent frameworks aim at bounded workflows: a task that runs once and exits. Division Swarm is for systems that run continuously, hold state across crashes, coordinate dozens of agents over days, and have to be auditable end to end. The technical separator: every transition commits as one transaction in a durable event store, the orchestrator never holds state in memory, and any run can be replayed or forked from the log. Use a workflow runner when the task finishes; use Division Swarm when the system has to last.
- Why YAML contracts? Why not a Python DSL?
- Because a contract bundle is read by a static analyzer before the runtime can boot. A Python DSL lets you write contracts the analyzer cannot see until the program runs, which defeats the fail-closed pitch. YAML is the smallest format the analyzer can fully verify offline, that tooling can render and generate, and that an LLM agent can author from a single machine-readable spec. The cost is verbosity; the benefit is that what passes verify cannot have whole classes of bugs that runtime-evaluated frameworks regularly ship.
- Is it production-ready?
- Pre-1.0. The platform specification is at v1.6.0, the engine ships in Phase 11, and the conformance suite covers 200+ contract bundles across 12 tiers. The runtime is used internally to operate autonomous multi-agent workflows; external production use is at your own risk and the API surface may change between minor versions until v1.0. The auditability primitives:event log, mutation log, replay, fork:are functional today.
- What does the deployment look like?
- A single Go binary, around 40 MB. Local and dev runs use file-backed SQLite at .swarm/dev.db with no database service required. Production opt-in is Postgres 16 via --store postgres. Agents that need the Docker workspace backend require the swarm-workspace image; everything else runs without containers. No Kubernetes required; a Linux box with a Postgres connection is enough.
- What does it cost?
- Apache 2.0 open source. Free to use, free to modify, free to operate in production. For teams running Division Swarm in regulated or high-stakes contexts, direct engineering partnership is available, scoped and time-boxed and led by the team that designed the runtime:see the integration-services section above.