CMMD-Center's local LLM serving infrastructure on Sentinel (Mac Studio) and Ryzen.
Apex is CMMD-Center’s local LLM serving stack. Mac Studio (Sentinel) runs the apex-* production tier; Ryzen runs the lab-* experimental tier. Both speak the Anthropic Messages API (/v1/messages) over Cloudflare tunnels and share a single Bearer key.
Endpoints
Section titled “Endpoints”| Endpoint | Tier | Host | Purpose |
|---|---|---|---|
https://apex-api.cmmd.ai | apex-* | Mac Studio (Sentinel) | Client-facing premium |
https://lab-apex-api.cmmd.ai | lab-* | Ryzen | Internal / experimental |
Deployed Models
Section titled “Deployed Models”| Alias | Endpoint | Backing model | Size | Context |
|---|---|---|---|---|
apex-core | apex-api | qwen3.5:122b-a10b | 81 GB | 262K |
lab-flash | lab-apex-api | GLM-4.6V Flash 9B | 6 GB | 128K |
lab-general | lab-apex-api | Qwen 3.6 27B (experimental) | 17 GB | 256K |
Source of truth: apex/models/registry.yml.
See Also
Section titled “See Also”- Architecture — Mac Studio + Ryzen + tunnel topology
- API Reference — request/response shape, auth
- Models — per-alias capability table
- Operations — runbooks, restart, debug
- Integration: CMMD — how CMMD calls Apex
- Integration: Forge — how Forge calls Apex
- Decisions — ADR-2026-05-19-apex-harness, ADR-2026-05-19-apex-model-choices
- Repo — infrastructure code