v0.6.66b485

Beta release. This is the server side of Ollama, LM Studio, and cloud model support in the Obsidian plugin (plugin v0.6.66b435).

Bring your own model runner: Ollama and LM Studio

lilbee can now use models served by a local Ollama or LM Studio endpoint, not just the GGUF models it downloads itself. Both are read-only: lilbee lists and runs their models but never pulls or deletes them, since their lifecycle is managed in those apps. Point lilbee at a running endpoint and its models join the catalog. A small per-backend spec (routing key, wire prefix, default URL, detection patterns) keeps adding another OpenAI-compatible runner cheap.

Cloud (API-key) models in the catalog

Models from providers you hold a key for — OpenAI, Anthropic, Gemini, and others — now appear in the model catalog next to native and local-server models, each tagged with its provider and whether the key is set (ready vs. missing key).

Granular model sources

The catalog, the installed-models API, and the CLI model list now report a precise source — native, frontier, ollama, or lm_studio — and a canonical prefixed reference (for example ollama/llama3). This is the contract the Obsidian plugin reads to group and label models by where they run.

TUI

The chat bar surfaces all four model roles: chat, embedding, vision, and rerank.

Full Changelog: v0.6.66b484...v0.6.66b485