3aef8678c0
Splits agent definition into stable provider + swappable model + role-bound profile. Adding a new LLM API is one row in providers.toml; new model is one row in models.toml; agent invocation picks any (provider, model) pair through agent-profiles.toml default_model_ref. - providers.toml: 10 providers — anthropic, openai, codex (OAuth), xai, deepseek, google, ollama-local, mlx-local, lmstudio-local, litellm-proxy - models.toml: 11 models with cost_*_per_mtok_micro + context_window + verified_at + deprecated_at - agent-profiles.toml: 18 representative profiles; manifest_path points to the canonical .md in ~/.claude/agents/ Three-layer DNA per the new architecture: agent-shell::<provider>:<model>:<caps>::<scope8>::<body8>-<nonce8> This commit only adds registries — kei-model-router still hardcodes the Claude-only Model enum. Wave 4 will rewire it to read TOML.
131 lines
4.1 KiB
TOML
131 lines
4.1 KiB
TOML
# Реестр провайдеров LLM. SSoT для kei-model-router + keisei-marketplace.
|
||
#
|
||
# Слой 1 из трёх: provider → model → profile (agent-profiles.toml).
|
||
# Провайдер стабилен (endpoint, схема аутентификации, биллинг-аккаунт),
|
||
# модели за ним меняются — см. models.toml.
|
||
#
|
||
# Все цены/контекст — в models.toml. Здесь только транспорт + лимиты.
|
||
#
|
||
# Конвенция: `id` — kebab-case, без версий. Версии — в моделях.
|
||
|
||
[[provider]]
|
||
id = "anthropic"
|
||
display_name = "Anthropic"
|
||
endpoint = "https://api.anthropic.com/v1/messages"
|
||
auth_scheme = "x-api-key" # header name; value from env ANTHROPIC_API_KEY
|
||
auth_env = "ANTHROPIC_API_KEY"
|
||
api_version_header = "anthropic-version"
|
||
api_version_value = "2023-06-01"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 4000
|
||
billing_currency = "USD"
|
||
notes = "Primary provider for Claude family. RULE 0.20 default for code-implementer-*."
|
||
|
||
[[provider]]
|
||
id = "openai"
|
||
display_name = "OpenAI"
|
||
endpoint = "https://api.openai.com/v1/chat/completions"
|
||
auth_scheme = "bearer"
|
||
auth_env = "OPENAI_API_KEY"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 5000
|
||
billing_currency = "USD"
|
||
notes = "GPT family + codex (ChatGPT OAuth, see codex provider for dual-review)."
|
||
|
||
[[provider]]
|
||
id = "codex"
|
||
display_name = "OpenAI Codex (ChatGPT OAuth)"
|
||
endpoint = "local:codex-cli" # invoked via `codex` CLI, not HTTP
|
||
auth_scheme = "oauth-subscription"
|
||
auth_env = "_" # no env — CLI handles auth
|
||
retry_max = 1
|
||
retry_backoff_ms = 0
|
||
rate_limit_rpm = 60 # ChatGPT subscription quota
|
||
billing_currency = "USD-subscription"
|
||
notes = "RULE 0.23 DUAL REVIEW. Separate billing from API providers."
|
||
|
||
[[provider]]
|
||
id = "xai"
|
||
display_name = "xAI"
|
||
endpoint = "https://api.x.ai/v1/chat/completions"
|
||
auth_scheme = "bearer"
|
||
auth_env = "XAI_API_KEY"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 1000
|
||
billing_currency = "USD"
|
||
notes = "Grok family. OpenAI-compatible API."
|
||
|
||
[[provider]]
|
||
id = "deepseek"
|
||
display_name = "DeepSeek"
|
||
endpoint = "https://api.deepseek.com/v1/chat/completions"
|
||
auth_scheme = "bearer"
|
||
auth_env = "DEEPSEEK_API_KEY"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 60
|
||
billing_currency = "USD" # invoices in USD even if upstream CNY
|
||
notes = "Cheapest tier for batch reasoning. R1 reasoner family."
|
||
|
||
[[provider]]
|
||
id = "google"
|
||
display_name = "Google Gemini"
|
||
endpoint = "https://generativelanguage.googleapis.com/v1beta/models"
|
||
auth_scheme = "query-key" # ?key=<...> in URL
|
||
auth_env = "GEMINI_API_KEY"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 1000
|
||
billing_currency = "USD"
|
||
notes = "1M-context window for long-doc analysis."
|
||
|
||
[[provider]]
|
||
id = "ollama-local"
|
||
display_name = "Ollama (local)"
|
||
endpoint = "http://127.0.0.1:11434/api/chat"
|
||
auth_scheme = "none"
|
||
auth_env = "_"
|
||
retry_max = 1
|
||
retry_backoff_ms = 0
|
||
rate_limit_rpm = 0 # local, no remote limit
|
||
billing_currency = "USD-zero" # local compute, opportunity cost only
|
||
notes = "Self-hosted. Models loaded via `ollama pull`."
|
||
|
||
[[provider]]
|
||
id = "mlx-local"
|
||
display_name = "MLX (Apple silicon local)"
|
||
endpoint = "http://127.0.0.1:8080/v1/chat/completions"
|
||
auth_scheme = "none"
|
||
auth_env = "_"
|
||
retry_max = 1
|
||
retry_backoff_ms = 0
|
||
rate_limit_rpm = 0
|
||
billing_currency = "USD-zero"
|
||
notes = "Native Apple-silicon inference. mlx_lm.server."
|
||
|
||
[[provider]]
|
||
id = "lmstudio-local"
|
||
display_name = "LM Studio (local)"
|
||
endpoint = "http://127.0.0.1:1234/v1/chat/completions"
|
||
auth_scheme = "none"
|
||
auth_env = "_"
|
||
retry_max = 1
|
||
retry_backoff_ms = 0
|
||
rate_limit_rpm = 0
|
||
billing_currency = "USD-zero"
|
||
notes = "Desktop GUI runner. OpenAI-compatible."
|
||
|
||
[[provider]]
|
||
id = "litellm-proxy"
|
||
display_name = "LiteLLM proxy (keisei.app)"
|
||
endpoint = "https://api.keisei.app/llm/v1/chat/completions"
|
||
auth_scheme = "bearer"
|
||
auth_env = "KEI_LITELLM_KEY"
|
||
retry_max = 3
|
||
retry_backoff_ms = 500
|
||
rate_limit_rpm = 1000
|
||
billing_currency = "USD-keisei"
|
||
notes = "Per-user-keyed proxy. Routes to anthropic/openai/xai/deepseek under one bearer."
|