PRODUCT
Deterministic automation for AI agents.
What Swamp is
You've seen what an agent can do. You've also seen it not do the same thing twice. Run the same operation and the second answer won't match the first. Now picture that's your SDLC, your release pipeline, or your calendar.
Swamp is the CLI your agent drives. Point it at any system and the agent reads the API docs, then writes a Swamp extension to talk to it.
Instead of performing the work each time, it builds a workflow. Typed models, secrets pulled from a vault, every result kept. The agent decides once; Swamp runs it the same way every time.
Local-first. Everything Swamp produces lands in .swamp on your disk. Models, run history, versioned data. Nothing leaves your environment unless you configure a datastore, and then it goes where you pointed it.
The stack
The pieces
Models
Typed connections to any API. The agent defines the shape; Swamp validates every call against it.
Workflows
Steps wired into one pipeline, built once and run on a schedule.
Data
Every run is saved — versioned, immutable, and there for the agent to read back next time.
Vaults
Encrypted secrets. The workflow names what it needs by key, and the value is injected at run time without ever entering a prompt.
Extensions
New capabilities the agent writes for itself. Publish one and every agent in the repo can pull it.
Reports
A markdown and JSON summary, written after every run.
Datastores
A shared home for .swamp, so run history follows your whole team instead of living on one laptop.
Swamp serve
A server that runs your workflows on demand, triggered by a webhook or a direct call from anywhere.