Decisions locked
the axes the fleet commits to| Axis | Decision | ADR |
|---|---|---|
| Primary model | Cloud is the default; local models never carry the main path. Talos, straylight, and titan primary is xAI grok-4.6. Straylight cloud fallback is zai glm-5.3-flash. | — |
| Local router | llama.cpp llama-server on straylight is the main-loop fallback/delegation endpoint. Aux routing (approval/titles/web_extract) is served by dixie:11434. | adr/0001-local-router.md |
| Transport | Mattermost is the gateway transport and hosts the 'claude' reviewer user. | — |
| Memory | Honcho runs on rift; every model role calls tether on dixie :11435 (upstream llama-server :11434) for gen, embed and dialectic. | adr/0002-memory-placement.md |
| GPU ceiling | Straylight (Strix Halo) holds models-max 3 under a ~104 GiB TTM cap. Titan holds models-max 2. Dixie holds two residents on 12 GB. | adr/0003-gpu-ceiling.md |
| Fetch | sift on orion is the fleet URL fetch (Firecrawl-compatible /v2/scrape and /v1/fetch) and search (native /v1/search and Firecrawl-compatible /v2/search). Search: Tavily first, SearXNG fill-in; Brave LLM available but disabled. | — |
| Dixie proxy | tether on dixie :11435 is the stream-converting OpenAI proxy in front of llama-server :11434. | — |
Structures
the 21 things that make up the fleetRepresentative flows
packet shapes the design impliesPayload shapes are what the design implies, not measured traffic. 7 flows walk the fleet.