anyagent

For apps that want to use the AI subscriptions a user already pays for.

One Up-to-date, feature-rich Rust interface to the agents already on a users machine.

Users have Claude Code, Codex, Cursor, Grok, OpenCode, Kiro, Antigravity, and friends on their
machines — each with its own CLI, protocol, and quirks. If you're building an
app on top of them, you end up writing and maintaining a driver per agent.
anyagent is that layer, once: it finds the agents, speaks each one's protocol
(native or ACP), and gives you one typed
API — Runtime, Session, and a stream of Events.

Your app keeps its own transcript, UI, and policy. anyagent owns the processes, the wires, and the turn rules — every agent behaves the same way through it.

Use

[dependencies]
anyagent = "0.0.1"

use anyagent::{EventKind, Runtime, SessionOptions};
use futures::StreamExt;

let runtime = Runtime::new();
let report = runtime.discover().await;
let agent = report.require("claude")?;
let (session, mut events) = runtime.open(agent, SessionOptions::in_dir(".")).await?;

session.prompt("explain this repo").await?;
while let Some(event) = events.next().await {
    match event?.kind {
        EventKind::TextDelta { text, .. } => print!("{text}"),
        EventKind::TurnEnded { .. } => break,
        _ => {}
    }
}
session.close().await?;

That's the whole core flow. Everything else is the same few objects:

Features

Capability	API
Discover installed agents and check login status	`runtime.discover()`, `runtime.probe(agent)`
Plan quota for the logged-in account	`runtime.plan_usage(agent)`
One-shot text (titles, commit messages) with no session to manage	`runtime.generate(agent, options, prompt)`
Send text, images, slash commands	`session.prompt(input)`
Steer or interrupt a running turn	`session.prompt(..)` mid-turn, `session.cancel()`
Answer a permission or question the agent raised	`session.answer(id, answer)`
Switch model / mode / any advertised option live	`session.configure("model", "sonnet")`
Resume, fork, or rewind a conversation	`SessionOptions::resume` / `fork_from`, `session.rollback(..)`
Add custom MCP Servers	`SessionOptions::mcp_server(..)`
Test your app against a scripted agent, no subprocess	`Runtime::with_mock(script)` (feature `mock`)

Events cover streamed text and reasoning, typed tool calls with diffs, plans,
token usage, permission requests, subagents, and turn boundaries — the same
EventKinds for every agent. Anything provider-specific rides in
extensions instead of leaking into the types.

Examples

examples/ is the tour — small, commented programs that run
against the agents on your machine:

chat.rs — the core loop: prompt, stream events,
answer permissions. Typing mid-turn steers the running turn.
sessions.rs — several sessions at once, then
resuming one by token.
probe.rs — discover installed agents and probe
their login state, models, commands, and capabilities.

Supported agents

Agent	Wire
Claude Code	native (stream-json)
Codex	native (app-server)
Antigravity	native (stream-json), or ACP through its optional server
pi	native (pi RPC)
opencode	native (HTTP + SSE)
Cursor CLI, Grok, Hermes Agent, Kiro CLI, Qwen Code	ACP

Any other ACP agent works without a catalog entry via
AgentInstallation::acp(name, path, args).

Docs

anyagent.mintlify.site

Using an AI agent to build on anyagent? Point it at one of these:

Source	Use
`docs/skill.md`	Agent skill: drop into `.claude/skills/anyagent/SKILL.md` (or your agent's skills dir)
llms.txt / llms-full.txt	Whole docs site as plain text
Context7	`use context7` in Cursor, Claude Code, etc.

Page	What it covers
Quickstart	Install, discover, open, prompt, stream
Core API	Runtime, Session, SessionOptions, every EventKind, requests, errors
Features	Each feature: what, how, what confirms it, what gates it
Agents	Catalog, capability matrix, per-agent quirks, custom ACP agents
Building an app	Step-by-step chat app: threads, permissions, settings, persistence, usage
Architecture	Layers, the engine/adapter seam, adding an agent

Generated API reference, every type and signature: docs.rs/anyagent.

Contributions

Contributions are welcome! Please be sure to open an issue first. Upon approval you may create a pr. Please avoid contributing AI-Slop :) ... unless fable wrote it. Read the contributing guide for the test tiers and what a PR needs.

License

MIT or Apache-2.0.

anyagent-rs