Tempest v0.1.17

Small release. Adds an Experimental features section in Settings, and the first thing living under it: warpLLM as an optional second chat agent inside canvas chat nodes, alongside Claude Code.

What's in this release

Experimental: warpLLM chat backend

warpLLM is an open-source Rust AI gateway — one SDK that routes to OpenAI, DeepSeek, Kimi, and OpenRouter via provider/model strings. We're integrating with it to support fellow open-source infra in the space; it ships as an in-process Rust dependency, not a sidecar, so there's no extra process to start or manage.

Turn it on under Settings → Experimental → Warp chat backend. A "Warp" group then appears in the chat node's Agents picker with a handful of models (GPT-5 Nano, GPT-4o Mini, DeepSeek V4, Kimi K3, Claude Sonnet 4 via OpenRouter). Off by default; nothing changes if you leave it off. Claude Code and BYOK chat are untouched either way.

Non-streaming for now — warpLLM has no streaming yet, so replies land as a single block. Keys are read from your process env (OPENAI_API_KEY, DEEPSEEK_API_KEY, MOONSHOT_API_KEY, OPENROUTER_API_KEY) with the matching BYOK slot as fallback.

Agents picker redesign

While wiring Warp in, the Agents category picker was flattened to one row per (agent, model) — same shape as the BYOK provider panel — and gained its own search box that matches on group name or model. The old inline model dropdown that was overlapping sibling rows is gone.

Platform Support

Windows requires WebView2 (bundled automatically when necessary).

Database Branches (from v0.1.5) additionally require Docker; without it the feature stays dormant and everything else works normally.

Atlas semantic search (from v0.1.10) downloads a ~25 MB embedding model on first opt-in.

What's next

Threads polish, Hephaestus hardening across all three OSes, Automations out of beta, and keyboard-native everything. warpLLM streaming and tool-use land when the upstream SDK does — until then the Warp backend stays experimental.