Question 1

How do I make multiple AI agents talk to each other?

Accepted Answer

llm-party connects multiple AI coding agents in a single terminal session. You type a message, tag an agent with @claude or @codex, and they respond in a shared conversation. Agents can also talk to each other directly using @next:tag to hand off work without you stepping in.

Question 2

Can I run Claude, Codex, and Copilot together in the same terminal?

Accepted Answer

Yes. llm-party runs Claude, Codex, Copilot, and any Claude-compatible API (Ollama, GLM, etc.) side by side in one terminal using official SDKs. All agents see the full conversation and can collaborate on the same codebase. No separate windows, no copy-pasting between tools.

Question 3

How do I delegate a task from one AI agent to another?

Accepted Answer

Agents in llm-party hand off work automatically using @next:tag. For example, you ask Claude to review code, Claude finds issues and passes the task to Codex with @next:codex, and Codex implements the fix. Up to 15 automatic handoffs per cycle, fully configurable.

Question 4

Is there an open source multi-agent AI orchestrator for the terminal?

Accepted Answer

llm-party is an open source, MIT-licensed multi-agent orchestrator that runs in your terminal. It connects AI coding agents as peers with no master/servant hierarchy. Install with bun add -g llm-party-cli and run llm-party in any project directory.

Question 5

What AI coding agents does llm-party support?

Accepted Answer

llm-party supports Claude (Anthropic Claude Agent SDK), Codex (OpenAI Codex SDK), and GitHub Copilot (Copilot SDK) as native providers. Any Claude-compatible API (Ollama, GLM, etc.) can be added as a custom provider with just a URL and auth token. All native integrations use official, published SDKs with no wrappers or output scraping.

Question 6

How do I use multiple AI coding agents at once?

Accepted Answer

Install llm-party with bun add -g llm-party-cli, then run llm-party in your project folder. The first run auto-detects which AI CLIs you have installed and sets up your config. Route messages with @tags like @claude, @codex, or @copilot. Use @all to broadcast to every agent at once.

Question 7

Do AI agents share context when working together in llm-party?

Accepted Answer

Yes. Every agent sees the full multi-party conversation in a shared context window. When Claude reviews code and Codex fixes it, Copilot can see both the review and the fix when you ask it to write tests. Sessions persist across messages and transcripts are saved as JSONL.

Question 8

What is the difference between llm-party and running AI agents in separate terminals?

Accepted Answer

Running agents in separate terminals means no shared context, no collaboration, and manual copy-pasting between tools. llm-party puts all agents in one session where they see each other's work, hand off tasks automatically, and build on each other's output. One terminal, shared conversation, zero copy-paste.

Question 9

Can I let Claude architect a solution and then have Codex implement it automatically?

Accepted Answer

Yes. Ask Claude to design the architecture, and Claude can hand off to Codex with @next:codex to implement it. Codex sees Claude's full design in the shared conversation and builds from it. You can then ask Copilot to write tests. The entire chain runs automatically with up to 15 hops.

Question 10

How do I switch between Anthropic and OpenAI models in the same terminal?

Accepted Answer

Route messages with @tags. Type @claude to talk to Anthropic's Claude, @codex to talk to OpenAI's Codex, or @copilot for GitHub Copilot. All models run in the same terminal session with shared context. No switching tools, no separate windows.

Question 11

Does llm-party work without MCP or a master-servant hierarchy?

Accepted Answer

Yes. llm-party uses a peer architecture where all agents are equals. No MCP protocol, no central controller, no master agent delegating to servants. You orchestrate from the terminal using @tags, and agents collaborate directly with each other through @next:tag handoffs.

Question 12

Can I save and resume AI agent sessions in llm-party?

Accepted Answer

Yes. llm-party saves session transcripts as JSONL files and a manifest with per-agent SDK session IDs and message cursors. Resume with llm-party --resume <session-id> or /resume in the TUI. Each agent picks up exactly where it left off with no duplicate messages. Works across all providers: Claude, Codex, and Copilot.

Question 13

What is the best way to orchestrate a team of AI agents from the command line?

Accepted Answer

llm-party is a CLI tool built for exactly this. Install with bun add -g llm-party-cli, configure your agents in JSON, and run llm-party in your project. Use @tags to route tasks, let agents hand off work to each other, and review the results. Native SDK integration means no wrappers or output scraping.

Question 14

Can I use llm-party for creative writing, storytelling, or roleplay?

Accepted Answer

Yes. llm-party is not limited to programming. Multiple agents can collaborate on stories, worldbuilding, roleplay scenarios, research, brainstorming, and any task where different perspectives improve the output. Agents maintain character, remember context across messages, and hand off narrative beats to each other.

Question 15

What are skills in llm-party and how do I use them?

Accepted Answer

Skills are folders containing a SKILL.md file with specialized instructions for specific workflows. Place them in ~/.llm-party/skills/ for global access or in your project's .llm-party/skills/ directory. Assign skills to specific agents using the preloadSkills config field. Agents load assigned skills on boot and use them when relevant.

Question 16

What is the mind-map in llm-party?

Accepted Answer

The mind-map is shared memory between all agents, stored as Obsidian-compatible markdown notes in ~/.llm-party/network/mind-map/. Agents write discoveries, constraints, session progress, and cross-project lessons as they work. Each entry uses wikilinks to connect related findings. Open the folder in Obsidian to visualize the knowledge graph across all your projects.

llm-party

Why throw a party?

Tag-Based Routing

Agent-to-Agent Handoff

Peer Architecture

Session Resume

Non-Blocking Queue

Direct SDK Integration

Extensible Adapters

Agent Sidebar

Cancel Panel

Skills

Mind-Map

Not Just for Programmers

Three steps to the party

Configure your agents

Launch the party

Talk to your team

The guest list

Claude

Codex

Copilot

Custom

Join the party