Getting started

Use cases

Roder is a harness, not just an end-user chat surface. The same runtime can sit behind a terminal client, a desktop app, a CI job, an eval runner, or a lab-specific agent distribution.

1. Build a coding-agent product

Compose the core with the provider, storage, policy, and tool extensions you want. Keep the runtime invariant layer stable while your product surface changes.

cargo run -p roder-cli
cargo run -p roder-cli -- app-server

Read next: App server, TUI, Desktop clients.

2. Build lab or eval infrastructure

Subscribe to typed events, persist the trajectory, score outcomes, and replay the run. Provider-specific wire formats are normalized before they reach the harness.

3. Add internal capabilities

Ship native Rust extensions for hot-path features, or attach MCP servers for language-independent tools. Both paths end up as tool specs and runtime events.

# ~/.roder/config.toml
[web_search]
enabled = true
provider = "tavily"

4. Run repeatable CI checks

Roder is early, but the workspace already supports normal Rust verification and deterministic mock-provider flows for offline harness tests.

name: roder-smoke

on:
  workflow_dispatch:

jobs:
  smoke:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: dtolnay/rust-toolchain@stable
      - run: cargo test --workspace
      - run: cargo test -p roder-cli

5. Orchestrate subagents

Enable the subagent extension, load disk-defined agents, and expose the canonical task tool so parent turns can delegate bounded work.

# ~/.roder/config.toml
[subagents]
enabled = true
default_agent = "explore"
max_concurrent = 2
max_depth = 1

6. Run domain-specific coding distributions

Generate headless builds that expose only the tools a task family should use. The zero-coder-edits profile combines the app server, disk context, process tasks, and Zerolang checked graph-edit tools for Zero source tasks.