agentdock vs OpenRath
TypeScript library giving a backend one agent runtime: session-scoped runs with streaming, approvals and cancel, a tool registry, injectable stores and a factory over nine AI SDK providers. — versus — PyTorch-shaped multi-agent framework: Session is the tensor, Agent the layer, Workflow the module — plus a v2 durable runtime with checkpoints, leases, an effect ledger and human interrupts.
Both answer 'embed an agent runtime in my own backend' rather than 'run a harness'. OpenRath is Python with PyTorch-style composition, a durable runtime, effect ledger and Postgres underneath; AgentDock is a TypeScript loop over the AI SDK small enough to read end to end. Opposite ends of the same decision — reach for OpenRath when runs must survive restarts, AgentDock when they must not add dependencies.
| agentdock | OpenRath | |
|---|---|---|
| Stars | 6 | 1.1k |
| Forks | 2 | 59 |
| Language | TypeScript | Python |
| License | MIT | BSD-3-Clause |
| Last activity | 3 days ago | 1 months ago |
| Topics | agents | agents, orchestration |
| Curated connections | 5 | 7 |
agentdock — the curator's take
A deliberately thin agent loop for Node backends. The one decision worth copying is that sessions own their message history through an injected store, so a multi-tenant app stops threading transcripts through every call site. Set expectations accordingly: v0.1.0, single-digit stars, in-memory defaults, and it is a wrapper over the Vercel AI SDK, so anything the SDK cannot do it cannot do either. Use it if you want a runtime small enough to read end to end; use a real harness if you want planning, sub-agents, sandboxes or memory.
OpenRath — the curator's take
The abstraction is the pitch. Starting from Session rather than the agent loop is what lets you branch, trace and compose many agents across many sessions without hand-rolling a state machine — and v2's effect ledger (idempotency keys, ambiguous effects parked in NEEDS_REVIEW instead of blindly replayed) is the part most 'production-ready' agent frameworks quietly skip. The cost is conceptual weight plus Postgres, Redis and S3 for anything durable. For one agent with five tools this is far more machinery than the job needs, and the Agent Server HTTP surface is still Beta.