agentd

An agent runtime. One static binary: give it an instruction and tools from remote MCP servers, and it runs the agentic loop — think, call a tool, observe, self-correct — as a one-shot job or a long-lived daemon.

$ cargo install agentd-cli
$ agentd --prompt "triage the newest issue and label it" \
    --mcp github=https://mcp-github.internal/mcp \
    --intelligence https://gateway.internal/v1

Or without Rust: curl -fsSL https://agentd.dev/install.sh | sh, or docker run --rm ghcr.io/agentd-dev/agentd:latest --help.

(The crate is agentd-cli because agentd on crates.io belongs to an unrelated project. The binary it installs is agentd.)

What it is

It runs no code of its own. Every capability is a tool on an MCP server you named. Nothing is implicit — what the agent can do is exactly what you wired.

The supervisor holds no model. Lifecycle, limits and the process tree belong to a component that cannot be prompted; the reasoning runs in child processes it can always kill. A jailbroken model still meets a process that will not negotiate.

It stays up. Durable workflows checkpoint before every effect, so a killed host resumes in-flight runs rather than starting over. A human step suspends a run and renders as an answerable question in every attached client — and the run survives a restart while it waits.

You can attach to it. agentd tui -c agent.yaml runs the daemon and a terminal UI together; there is a web UI too. The daemon holds all the state, so quitting a client leaves the agent working.

In 20 lines

config_version: "2"
agent:
  name: triage
  instruction: You triage incoming issues. Be precise; ask if unsure.
intelligence:
  endpoints: https://gateway.internal/v1
  model: gpt-5.1
mcp:
  servers:
    - { name: github, endpoint: https://mcp-github.internal/mcp }
store: { kind: mcp, mcp: { server: github } }
workflows:
  - name: triage
    steps:
      new:  { kind: subscribe, server: github, uri: "github:///issues" }
      work: { kind: agent, depends_on: [new], instruction: "Triage {{payload.title}}" }
      done: { kind: finish, depends_on: [work], status: completed }

agentd --validate-config -c agent.yaml checks it — including the workflow bodies — before anything runs.

Documentation

https://agentd.dev/docs/ — getting started, configuration, workflows, MCP, A2A, security, deployment. Licensed under Apache-2.0.