| name: | imperial-free-llm-router |
|---|
| description: | Use the Val Town included free OpenAI-compatible proxy as a bounded advisory LLM for IMPERIAL agents. Enforces quota, no-spend mode, no tool authority, no signing, no transfer, JSON contracts, audit hashes, and fail-closed validation. |
|---|
Provider path: Val Town standard OpenAI-compatible free proxy.
Policy:
- FREE_ONLY. No card, trial, paid fallback, billing activation, or external paid API key.
- Advisory output only. LLM output never grants capabilities, approvals, signing, wallet control, outreach authority, or financial execution.
- Daily quota and minimum call interval are enforced in SQLite.
- Treat model output as untrusted. Require strict JSON contract and hash every prompt/response.
- Current live candidate model:
gpt-5.6-luna; runtime verification must confirm the returned model before status can be LIVE_ADVISORY_VERIFIED.
- DICON may use the model for offer design, proposal drafting, CRM analysis, research planning and internal next-best-action decisions.
- External messaging remains T2 approval-gated. Contracts, payment destinations and transfers remain T3/financial approval-gated.