# 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.

- Skill: `ydf0509/local-agent-model-defaults` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add ydf0509/local-agent-model-defaults`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ydf0509/local-agent-model-defaults/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: ydf0509 (https://skillmd.com/u/ydf0509)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ydf0509/local-agent-model-defaults

---


# 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`:

```dotenv
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:

```python
from langchain_openai import ChatOpenAI

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