AGENT SKILL

local-agent-model-defaults

Use when creating or modifying AI agent projects and choosing a model, API key, OpenAI-compatible base URL, LiteLLM endpoint, or related environment variables.

License 许可
MIT
所属仓库
ydf0509/lc-agent
最近更新
2026-11-05
Stars
★ 8
Forks
2

01 前言 Front Matter

name
local-agent-model-defaults
description
Use when creating or modifying AI agent projects and choosing a model, API key, OpenAI-compatible base URL, LiteLLM endpoint, or related environment variables.

02 资源文件 1

03 SKILL.md 正文

Local Agent Model Defaults

Use these personal defaults for AI agent projects unless the user or the project explicitly specifies another provider, model, API key, or endpoint.

Local LiteLLM Defaults

Prefer the local OpenAI-compatible LiteLLM proxy on port 4000:

OPENAI_API_KEY=sk-no-key-needed
OPENAI_BASE_URL=http://localhost:4000/v1
OPENAI_MODEL=ds-deepseek-v4-flash

The OpenAI-compatible client uses http://localhost:4000/v1 as its base URL and sends chat requests to /chat/completions. The local proxy does not require a real provider API key, so sk-no-key-needed is the intended placeholder.

In lc-agent's config.jsonc, the same model name goes into BOTH required fields of a model entry: model_id (your own globally-unique name — identity, stats, display) and raw_model_id (the channel-provided original model name — pricing fallback + stats merging). Under this LiteLLM proxy both values are identical, e.g. {"model_id": "ds-deepseek-v4-flash", "raw_model_id": "ds-deepseek-v4-flash"}. Missing fields or uniqueness conflicts (model_id global-unique; (provider, raw_model_id) unique) fail at startup with a full problem list.

Decision Rules

  1. Use the local LiteLLM values above when an AI agent implementation needs model configuration and no explicit alternative is requested.
  2. Preserve an explicit model, provider, base URL, or secret supplied by the user or already required by the project.
  3. Never invent or request a real external API key when the local LiteLLM proxy is the intended backend.
  4. Keep secrets out of source control; use environment variables or the project's existing configuration mechanism.

Example

For a new OpenAI-compatible agent configuration, use:

from langchain_openai import ChatOpenAI

model = ChatOpenAI(
    model="ds-deepseek-v4-flash",
    api_key="sk-no-key-needed",
    base_url="http://localhost:4000/v1",
)

04 同仓库更多 Skills