9b41d51231
Source snapshot: 514ab1975951d94342ea38e64101d5a0f1c51c77
1.5 KiB
AI
Tolaria has two AI paths: coding agents that can use tools to inspect and edit a vault, and direct model targets that answer in chat mode from note context.
Coding Agents
The AI panel can stream supported local CLI agents through Tolaria's normalized event layer. Current targets include Claude Code, Codex, OpenCode, Pi, and Antigravity CLI when they are installed on the machine.
Coding agents can run in:
- Vault Safe mode, limited to file, search, and edit tools.
- Power User mode, which can allow local shell commands scoped to the active vault for agents that support shell access.
Direct Models
Direct model targets run in chat mode. They receive the active note, linked context, and conversation history, but they do not receive vault-write tools or shell access.
Supported provider shapes include:
- Local models through Ollama or LM Studio.
- Hosted providers such as OpenAI, Anthropic, Gemini, and OpenRouter.
- Custom OpenAI-compatible endpoints.
External MCP Setup
Tolaria exposes an MCP server for external tools. The setup flow can write Tolaria's MCP entry into Claude Code, Antigravity CLI, Cursor, and a generic MCP config path, and it can also copy the exact JSON snippet for manual setup.
MCP setup is explicit. Closing the dialog leaves third-party config files untouched.
Why Git Matters For AI
AI-generated changes should be inspectable. Git gives you diffs, history, rollback, and a clear boundary between suggestions and committed work.