Download
Features Pricing Manifesto

AI

Local models

Run models on your machine and point Glyph at them. Local setups use the same Glyph AI panel once AI features are on. Configure them under Settings → Glyph AI for the open Space.

Local endpoints are per space, same as API keys. Allow local network must be on for localhost or LAN servers.

Ollama

Ollama defaults to localhost:11434.

  1. Install and start Ollama on your computer.
  2. Pull the models you want in Ollama.
  3. In Glyph, select Ollama (or the matching local option) and confirm the endpoint.
  4. Turn on Allow local network so Glyph can reach the local server.
  5. Enable AI features and open the Glyph AI panel.

Keep Ollama running while you chat. If Ollama is stopped, Send fails until you start it again.

llama.cpp

llama.cpp servers default to localhost:8080/v1.

  1. Start your llama.cpp server with an OpenAI-compatible HTTP API.
  2. In Glyph, point at that base URL (default http://localhost:8080/v1 unless you changed the port).
  3. Enable Allow local network.
  4. Send a test message from the Glyph AI panel.

Match the port in Glyph to the port your server prints when it starts.

OpenAI compatible (LM Studio and similar)

Apps like LM Studio often expose an OpenAI-compatible API.

  1. Start the local server in that app and note the Base URL it shows.
  2. In Glyph, choose OpenAI compatible.
  3. Set the Base URL to that address.
  4. Enable Allow local network.
  5. Set a key only if the local app requires one.

You may not need a real API key for a local server. Follow what your local app documents. Glyph still stores the connection with the Space.

Allow local network

Allow local network must be on for Glyph to talk to servers on localhost or your LAN. Leave it off when you only use remote cloud APIs.

Turn it on for Ollama, llama.cpp, LM Studio, and other local OpenAI-compatible servers. Turn it off if you switched back to cloud-only keys and want local access disabled.

How this fits with other AI options

  • Local models: this guide (Ollama, llama.cpp, OpenAI compatible local).
  • Cloud API keys: OpenAI, Anthropic, Google/Gemini, OpenRouter.
  • Agents: Codex, OpenCode, Amp, Claude Code, PI with their CLIs.

Pick one approach per Space profile, or keep several configured and switch which one you use in the panel. Chat, tools, and @ context follow the Glyph AI overview.

Tips

  • If chat fails, confirm the local server is running and the port matches Glyph.
  • Large models need RAM and time; a hung Send often means the server is still loading weights.
  • Tools on vs off still applies: chat-only vs create mode is independent of where the model runs.