Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio + HTTP/JSON (hosted relay, staging)
Certification scope
PASS (hosted/staging) — Verity delta re-bench, 2026-10-27 (llm-connectors-20261027/bench-groq.md): adapter 0.1.1 on core 0.1.2 (regression on core 0.1.2 PASS in bench-anthropic.md delta re-bench 2); conditions C1–C4 all CLEARED, no surviving conditions. PASS is not CERTIFIED — certified is false, and nothing beyond this verdict at hosted/staging scope is claimed. Hosted deployment is bench-passed at hosted/staging and is NOT yet public-live. Catalogue is exactly openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b (PREVIEW) — the gpt-oss route, never 'all open weights'. Groq is a trademark of Groq LLC and/or its affiliates. Eclipse Connect for Groq is not affiliated with or endorsed by Groq LLC.
What it does
- BYOK custody on the Cyan Eclipse Connect LLM core (core 0.1.2, PASS (hosted/staging)): your own provider API key is held only by the relay, encrypted (AES-256-GCM under a master key, via the core vault), under a signed, scoped, revocable grant — model allowlist, spend cap, expiry — and the agent never touches the key
- Every call is pre-checked against the grant and sealed into a content-free Herald-signed receipt; receipts are independently verifiable off-host from the published JWKS surface
- Instant kill switch: delete the key and kill every grant instantly; deleting a key cascade-revokes its grants; receipts are retained (they are content-free)
- Groq OpenAI-compatible transport with Bearer auth injection on the grant's own key only; catalogue exactly openai/gpt-oss-120b ($0.15/$0.60 per 1M), openai/gpt-oss-20b ($0.075/$0.30) and qwen/qwen3.8-27b ($0.80/$4.00, PREVIEW) — price table groq-prices/2026-10-27 as of 2026-10-27; this is the gpt-oss route, never 'all open weights'
- Disclosure (SPEC §11.4(i)): you remain the provider's customer and are liable for your own account activity and bills; provider token costs are billed by the provider directly to you
- Disclosure (SPEC §11.4(ii)): the relay holds your key encrypted; the key is deletable and every grant is killable instantly
- Disclosure (SPEC §11.4(iii)): content logging is OFF by default; it can be turned on only per grant by you, never by an agent, and receipts themselves contain no prompt or response content, ever
- Disclosure (SPEC §11.4(iv), Groq): Groq rate limits are per-organisation, not per-key; qwen/qwen3.8-27b carries a PREVIEW label at Groq
- Trademark: 'Groq' is used as a word mark only (nominative fair use), no Groq logo. Groq is a trademark of Groq LLC and/or its affiliates. Cyan Eclipse Connect for Groq is not affiliated with or endorsed by Groq LLC.
Tools
llm_callllm_modelsllm_balancellm_receiptsllm_receipt_verify
Problems it addresses
- I need my agent to call Groq's gpt-oss models without ever holding my Groq API key
- I need Groq calls bounded by a model allowlist, a spend cap, and an expiry — with a sealed receipt for every call
- I need the gpt-oss route on Groq with per-organisation limits stated honestly, not per-key promises
Not tested
- Any live Groq call — bench ran against recorded fixtures/mocks only; no live provider smoke at this scope
- Public-live hosted deployment — bench scope is hosted/staging only
Evaluation-intake listing
This listing records the tool's place in the Bazaar's evaluation
and monitoring pipeline, with the evidence gathered so far printed above — verdict, scope,
and not-tested items included. It is not sold on this site, and its presence in the catalog
is not a claim beyond that evidence. The service behind every verdict is the bench:
what a bench run buys →