Models

File: config/agents/models.yaml · Schema: schemas/models.schema.json

The model registry: which provider serves each model, its pinned revision and its license, checked against the allow-list in policies.yaml.

Fields

Field	Type	Description
allowed_licenses	list of string	Required. SPDX ids a model may be licensed under. A model whose license is not listed stops boot.
models	map of name → ModelDef	Required. Models by key; llm.extract and records.match name a model by this key.

Unknown fields are refused at boot.

Types

Types referred to by the fields above, from the same schema.

ModelDef

Field	Type	Description
dimensions	integer (uint32) (optional)	Vector width; required for an embedding model, and must equal the embeddings.embedding column width.
license	string	Required. SPDX license id; must be in allowed_licenses.
name	string	Required. The model name the provider serves, e.g. Qwen/Qwen3.8-27B-FP8.
params	any	Request parameters sent with every completion: temperature, max_tokens, enable_thinking, ….
provider	string	Required. Key in settings.llm.providers.
revision	string	Required. The exact revision approved for use (a Hugging Face commit hash). Recorded here; nothing reads it at run time yet.
role	string	Required. chat (used by llm.extract) or embedding (used by ingestion and records.match).

Unknown fields are refused at boot.

In this repository

config/agents/models.yaml

# The model catalogue. Boot fails if any model's license is not in allowed_licenses.
models:
  chat_default:
    provider: vllm_chat # key in settings.llm.providers
    name: Qwen/Qwen3.8-27B-FP8
    revision: "<hf-commit-hash>" # pin before production (plan §18)
    license: Apache-2.0
    role: chat
    params: { temperature: 0, max_tokens: 4096, enable_thinking: false }
  embed_default:
    provider: vllm_embed
    name: Qwen/Qwen3-Embedding-0.6B
    revision: "<hf-commit-hash>"
    license: Apache-2.0
    role: embedding
    dimensions: 1024 # must equal the embeddings.embedding column width
allowed_licenses: [Apache-2.0, MIT, BSD-3-Clause]

Editor support

The schemas are generated from the Rust definition structs (task schema_export), and a test fails when they drift. Point a YAML language server at the schema to get completion and validation while editing:

# yaml-language-server: $schema=../../schemas/models.schema.json