# llm-party

> Peer orchestrator that puts multiple AI agents in the same terminal. You talk, they listen. They talk to each other. Nobody is the boss except you.

## The Problem

Developers juggle multiple AI tools in separate windows. Claude in one terminal, Codex in another, Copilot somewhere else. No shared context. No collaboration. Copy-paste is the "integration."

## The Solution

llm-party connects Claude, Codex, Copilot, and any Claude-compatible API (Ollama, GLM, etc.) in a single terminal session. Every agent sees the full conversation. They hand off work to each other automatically. You route with @tags.

```
YOU > @claude review this function
[CLAUDE] The error handling on line 42 silently swallows exceptions...

YOU > @codex fix what claude found
[CODEX] Fixed. Added proper error propagation with context.

YOU > @copilot write tests for the fix
[COPILOT] Added 3 test cases covering the new error paths. All passing.
```

## How It Works

- Peer architecture: no master/servant hierarchy, no MCP. All agents are equals.
- Direct SDK integration: official published SDKs from each provider. Nothing reverse-engineered, patched, or bypassed.
- Persistent sessions: agents maintain context across messages. Resume with --resume or /resume. Per-agent SDK session IDs and message cursors preserved in a manifest file.
- Non-blocking queue: each agent has its own processing queue. You can type while agents work. Fast agents respond immediately, slow agents queue and process when ready.
- Shared context: every agent sees only unseen messages since its last turn. No duplicate processing on resume or concurrent dispatch.
- Agent-to-agent handoff: agents pass work to each other with @next:tag. Up to 15 automatic hops per cycle.
- Tag-based routing: @claude, @codex, @copilot, @all. Messages go exactly where you want.
- Agent sidebar: live activity panel showing what each agent is doing (file reads, shell commands, search queries). Toggle with Ctrl+B.
- Cancel panel: press Esc to selectively cancel running agents while others keep working.
- Animated splash screen: octopus mascot displayed when the terminal is idle.

## SDK Transparency

llm-party uses official, publicly available SDKs. All authentication flows through each provider's own CLI login. No credentials are stored, proxied, or intercepted.

| Provider | Official SDK | Published by |
| -------- | --------------------------------- | ------------ |
| Claude | @anthropic-ai/claude-agent-sdk | Anthropic |
| Codex | @openai/codex-sdk | OpenAI |
| Copilot | @github/copilot-sdk | GitHub |
| Custom | Any Claude-compatible API (Ollama, GLM, etc.) | Via native CLI |

## Install

Requires Bun runtime and at least one AI CLI installed and authenticated (claude, codex, or copilot).

```bash
bun add -g llm-party-cli
```

Run inside any project directory:

```bash
your-project/> llm-party
```

First run launches an interactive setup wizard that auto-detects installed CLIs. Custom providers (Ollama, GLM, etc.) can be added with just a URL and auth token. If your AI CLIs work on their own, they work inside llm-party.

## Configuration

Edit ~/.llm-party/config.json to add or change agents:

```json
{
  "agents": [
    { "name": "Claude", "tag": "claude", "provider": "claude", "model": "opus" },
    { "name": "Codex", "tag": "codex", "provider": "codex", "model": "gpt-5.2" },
    { "name": "Copilot", "tag": "copilot", "provider": "copilot", "model": "gpt-4.1" }
  ]
}
```

That's it. Name, tag, provider, model. No paths, no prompts, no usernames to configure.

## Safety Notice

All agents run with full permissions. They can read, write, edit files and execute shell commands with zero approval gates. Always run in a disposable environment: a Docker container, a VM, or at minimum a throwaway git branch.

## Quick Facts

- Package: llm-party-cli on npm
- License: MIT
- Runtime: Bun
- Built by: AALA Solutions (aalasolutions.com)
- Repository: github.com/aalasolutions/llm-party
- Website: llm-party.party