Skip to content

getting-started.md

Latest commit

 

History

History
98 lines (79 loc) · 3.11 KB

getting-started.md

File metadata and controls

98 lines (79 loc) · 3.11 KB
Raw

getting started

Quick start

  1. set OPENROUTER_API_KEY in .env
  2. optionally set OPENROUTER_MODEL
  3. run one of the examples

Examples:

  • cargo run -p openrouter-chat -- "hello"
  • cargo run -p openrouter-coding-agent -- "Use fs_read_file on ./Cargo.toml and return only the workspace member count as an integer."
  • cargo run -p openrouter-agent-cli -- --mcp-mock "Return only the secret from the MCP tool."

To run the Anthropic example instead, set ANTHROPIC_API_KEY, ANTHROPIC_MODEL, and ANTHROPIC_MAX_TOKENS (required by the Messages API), then:

  • cargo run -p anthropic-chat -- --web-search 3 --thinking 2048

To run the Cerebras examples, set CEREBRAS_API_KEY and CEREBRAS_MODEL (both required; CEREBRAS_BASE_URL, CEREBRAS_VERSION_PATCH, and CEREBRAS_MAX_COMPLETION_TOKENS are optional), then:

  • cargo run -p cerebras-chat -- --reasoning-effort medium
  • cargo run -p cerebras-batch -- run ./prompts.json

Minimal composition

The smallest useful assembly is:

  • one model adapter
  • one tool registry
  • one permission checker
  • one loop observer
let agent = Agent::builder()
    .model(adapter)
    .tools(agentkit_tool_fs::registry())
    .permissions(my_permissions)
    .observer(my_reporter)
    .build()?;

Then:

let mut driver = agent
    .start(SessionConfig {
        session_id: SessionId::new("demo"),
        metadata: MetadataMap::new(),
    })
    .await?;

driver.submit_input(vec![system_item, user_item])?;

match driver.next().await? {
    LoopStep::Finished(result) => { /* render output */ }
    LoopStep::Interrupt(interrupt) => { /* approval, auth, or input */ }
}

Example progression

The examples are meant to build up in complexity:

  • openrouter-chat
    • provider + loop
  • openrouter-coding-agent
    • provider + loop + fs tools + permissions
  • openrouter-context-agent
    • provider + loop + context loading
  • openrouter-mcp-tool
    • provider + loop + MCP tool adaptation
  • openrouter-subagent-tool
    • custom tool extension
  • openrouter-acp-trio
    • three agents exposed as in-memory ACP endpoints, delegating to each other over the Agent Client Protocol
  • openrouter-compaction-agent
    • structural, semantic, and hybrid compaction with a nested-loop compaction backend
  • openrouter-agent-cli
    • combined example: context + tools + shell + MCP + compaction + reporting
  • anthropic-chat
    • interactive REPL against Anthropic's Messages API, exercising streaming, server tools (web search, web fetch, code execution), extended thinking, and the buffered/streaming toggle
  • cerebras-chat
    • interactive REPL against Cerebras' /v1/chat/completions, exposing every CerebrasConfig knob (sampling, reasoning, response format, compression, service tier, predicted outputs, local + MCP tools) as CLI flags, with slash commands (/show, /usage, /ratelimit, /headers, /models, /reset) for inspecting runtime state
  • cerebras-batch
    • one-shot CLI over the Files + Batch APIs: files upload|list|get|content|delete, batches create|submit|list|get|cancel|wait, and run for the submit → wait → fetch happy path