DocsSupported Agents

Supported Agents

Blueprint Stack ships with 4 built-in agents — two cloud coding agents (Claude Code, Codex) and two local options that run open models entirely on your own machine (Ollama, LM Studio) — plus unlimited custom agents. You can switch between agents per session or run several in parallel across tabs; selecting a different agent from the picker opens a new tab automatically. Blueprint Stack does not auto-download dependencies: cloud agents need their CLI / adapter installed, and the local agents need their server app installed and running. Select an agent below for setup instructions.

Cloud vs. local

  • -Cloud agents (Claude Code, Codex) run on the provider's models. They need the agent CLI / adapter installed and an account or API key, and your prompts go to the provider.
  • -Local agents (Ollama, LM Studio) run open models on your hardware. No API key, no Node.js, no adapter to install — just the Ollama / LM Studio app with its server running and a model loaded. Once the model is downloaded, they work fully offline, and nothing leaves your machine. Agent quality depends on the model and hardware you choose, and the model should support tool / function calling so it can drive the editor.

Agent Picker Behavior

  • -Selecting a different agent from the picker opens a new tab for that agent.
  • -Your current chat tab is preserved, so you can keep multiple concurrent sessions (for example, one Claude Code tab and one Codex tab).
  • -Each agent in the picker shows a status label: Available, Not Installed, or Missing API Key.

Built-in Agents

Claude Code

RecommendedCloud

Anthropic's coding agent. Connects through the claude-code-acp adapter. Supports session resume, extended thinking, and streaming.

Models:Claude Sonnet 4, Claude Opus 4
# Quick setup
npm install -g @zed-industries/claude-code-acp

Codex (OpenAI)

Cloud

OpenAI's Codex agent. Requires the codex-acp adapter for Blueprint Stack integration.

Models:GPT-4, o3, o4-mini
# Quick setup
npm install -g @openai/codex && npm install -g @zed-industries/codex-acp

Local (Ollama)

Local / offline

Run open models locally with Ollama — no API key, no Node.js, no extra adapter, and nothing leaves your machine. Pick a tool-calling-capable model (Llama 3.1, Qwen2.5, etc.).

Models:Any model in your Ollama library (tool-calling models recommended)
# Quick setup
Install Ollama from ollama.com, then `ollama pull llama3.1`

Local (LM Studio)

Local / offline

Run open models locally through LM Studio's built-in server. Fully on-device — no API key, no Node.js, no adapter. Load a model that supports tool / function calling.

Models:Whatever you load in LM Studio (tool-calling models recommended)
# Quick setup
Install LM Studio from lmstudio.ai, then Developer tab → Start Server

Custom Agents

You can add unlimited custom ACP-compatible agents via Project Settings > Plugins > Blueprint Stack > Agents.

Custom agent configuration includes:

  • -Executable path - Path to the agent binary or command name.
  • -Arguments - Command-line arguments passed to the executable.
  • -Working directory - Directory the agent process runs in.
  • -Environment variables - Key-value pairs injected into the agent process.
  • -API key - Optional authentication credential.
  • -Model ID - Optional model identifier to pass to the agent.
  • -Install command - Shell command shown to the user for installing the agent.

Path Resolution

For cloud and custom agents that run as a command, Blueprint Stack resolves the executable by searching the following locations (local agents like Ollama and LM Studio have nothing to resolve — they connect to a server address you can configure in settings):

  • -System PATH
  • -npm global bin directory
  • -bun global bin directory
  • -cargo bin directory
  • -volta bin directory
  • -fnm bin directory
  • -asdf shims directory
  • -mise shims directory
  • -scoop shims directory
  • -Homebrew bin directory

Status Labels

The agent picker shows a live status for each agent:

  • -Available - The agent executable was found and is ready to use.
  • -Not Installed - The agent executable was not found in any resolved path.
  • -Missing API Key - The executable was found but authentication is not configured.

Local agents (Ollama, LM Studio) always show Available — there's nothing to install or authenticate. Whether the local server is actually running is checked the first time you send a prompt; if it isn't, you'll get a clear “couldn't reach” message telling you to start it.