Handbook
LLM gateway (forge-llm)
forge-llm is an optional local AI lab: Docker Compose gateway (Ollama-compatible), ComfyUI, and Prometheus telemetry. It is not a governed task runner — that is LCDL.
Updated
When to use forge-llm
Use when operators want a self-hosted OpenAI-compatible endpoint for LCDL LLM_* profiles or Lenses Ollama settings — without routing inference through cloud APIs.
Default surfaces
| Service | Typical port | Notes |
|---|---|---|
| Gateway | 11434 |
Ollama-compatible HTTP |
| Console | 8787 |
Operator UI |
| Bind | HOST_BIND_IP |
See forge-llm/docs/OPERATIONS.md |
Integrations
| Consumer | Contract |
|---|---|
| LCDL | LLM_* env profile points at gateway; gateway probes in LCDL handbook |
| Lenses | LLM settings may target local Ollama-compatible endpoint |
| forge-workcells | local_llm_worker uses same LLM_* profile as KS harness --llm path |
| Fleet (legacy) | Some installs expose /v1/services/forge-llm/* — prefer direct gateway for new work |
See ecosystem reference micro-agent flow and intersection map.
Does not own
forge-llm does not own governed contracts, Lenses SoR, or platform policy. LCDL remains the task and trace layer.
In-repo documentation
forge-llm/README.md and docs/OPERATIONS.md — GPU mode, model presets, compose stack.
No Firebase handbook shell. Platform link-out only.