All plugins

kimchi-provider

Kimchi (kimchi.dev) models via the OpenAI-compatible API — live catalog from the metadata endpoint, API-key auth, gateway unknown-model 400s classified as model_not_found, and an identifying User-Agent (hermes-agent/<version>) on every gateway request.

getkimchi1Modelsv1.1.0Updated 1 month ago

About

Kimchi (kimchi.dev) models via the OpenAI-compatible API — live catalog from the metadata endpoint, API-key auth, gateway unknown-model 400s classified as model_not_found, and an identifying User-Agent (hermes-agent/<version>) on every gateway request.

kimchi-provider — Kimchi models via API key

Hermes Agent model-provider plugin: use Kimchi
models with Hermes' own agent loop over the OpenAI-compatible gateway
(https://llm.kimchi.dev/openai/v1).

This is Layer 1 of the two-plugin
kimchi-hermes monorepo —
the API-key alternative to
kimchi-acp-provider, which drives the full Kimchi
harness over ACP.

Install

hermes plugins install getkimchi/kimchi-hermes/model-providers/kimchi
# or, once listed in the Hermes plugin catalog:
hermes plugins install kimchi-provider

The installer prompts for KIMCHI_API_KEY if it is not already set
(the key is stored in ~/.hermes/.env, masked at entry).

Authenticate

export KIMCHI_API_KEY=...   # or persist it in ~/.hermes/.env

Use

hermes model    # pick "Kimchi"
hermes          # you're on Kimchi

Or make it the default in ~/.hermes/config.yaml:

model:
  default: glm-5.3-flash
  provider: kimchi

The model picker lists the live catalog fetched from Kimchi's model
metadata endpoint. There is deliberately no hardcoded fallback list: a
transient fetch failure means an empty picker until the next attempt —
never a stale list.

Environment variables

VariablePurpose
KIMCHI_API_KEYAPI key (required; checked before ~/.hermes/.env)
KIMCHI_BASE_URLOverride the inference gateway base URL — must include the /openai/v1 path (e.g. https://llm.eu.kimchi.dev/openai/v1). Caution: the Kimchi harness reads the same-named variable with bare gateway-base semantics (no path, it appends /openai/v1 itself); avoid exporting it shell-wide when you also use kimchi-acp — see the monorepo README env table

Behaviour & disclosures

  • Network: fetches the live model catalog from
    https://llm.kimchi.dev/v1/models/metadata with your API key
    attached (Authorization header, via Hermes' credentialed-URL
    wrapper). Model inference goes through Hermes' own provider client
    to https://llm.kimchi.dev/openai/v1.
  • Credentials: none stored by the plugin — the key is resolved by
    Hermes' built-in env / ~/.hermes/.env ladder.
  • Error mapping: the Kimchi gateway's
    "no registered providers found" 400 (unknown model id) is classified
    as model_not_found, so Hermes' model-not-found recovery applies
    instead of a generic format error.
  • Registers only a provider profile (register_provider); no tools,
    no hooks, no middleware.

Known limitations

See the monorepo
README
(Roadmap & known gaps) and SPEC.md for the full
evidence table and decision log.

README from the reviewed commit 0fcca81; it updates when the author re-pins.

More by getkimchi

2 plugins by getkimchi in the catalog.

Developer Mode

Explore built-in Hermes toolsets and actions.

HermesAgent

Developer?

Create a Hermes extension for Desktop or agent workflows.

Developer Section