LLM Chat
The LLM Chat pane lets an AI agent edit your timeline, media and transcripts. Try asking “remove silence from my selected clip”.
Setup
Open Settings -> AI Providers, then install and sign in to a provider.
Providers
| Provider | Install | Sign in |
|---|---|---|
| Codex | developers.openai.com/codex/cli | codex login |
| OpenCode | opencode.ai | opencode auth login |
| Grok | x.ai/cli | grok login |
| Lemonade | Built in | No sign in, runs locally |
| Antigravity | Install button in settings | Sign in button in settings |
Lemonade runs models on your PC. Download one from Settings -> AI Providers -> Download models, or set a Server URL to use your own Lemonade server. It also powers Transcript search, letting the AI find topics across all transcripts by meaning.
Claude can’t run as an in-app provider. Run claude in the chat pane’s terminal instead, or connect it as an external client.
Note: Agents run with full access and auto-approve their actions. Clumper tells them to only use Clumper’s tools.
Chatting
- Pick a model from the searchable model picker. The model is locked after the first message; start a New Chat (Ctrl + Shift + O) to switch.
- Selected clips and transcript text show as badges and are sent to the agent as context.
- Messages sent while the agent is working are queued at the bottom. Stopping returns them to the composer.
- Tool calls are grouped and can be collapsed. Click Undo these edits to revert a reply’s changes.
- Chat is unavailable inside Remove Silence or preview mode.
Terminal
The pane includes a terminal. Set a Launch command to run automatically each time it opens (e.g. claude).
External clients
Enable Allow external AI clients in settings to let tools like Claude Code edit your project through Clumper’s local MCP server. Clients must send the access token shown in settings. Keep Clumper open while they work.