---
name: llm
description: >
  LLM provider integrations for 20+ services including OpenAI, Anthropic, Google, Ollama, Groq, Mistral, and more, with unified streaming chat completion and model management APIs.
metadata:
  author: ysnows
  version: "1.3.377"
---

## Architecture

### Unified Streaming Architecture
All providers convert their native stream format into an **Anthropic-like `Stream<BaseChatMessageChunk>`** format. The unified consumer in `enconvo.nodejs` (`LLMProvider.handleAgentMessages()`) processes these normalized chunks identically.

Stream format contract per content block: `content_block_start` → `content_block_delta`* → `content_block_stop`, plus a final `usage` chunk.

Delta types: `text_delta` (text), `thinking_delta` (reasoning), `signature_delta` (thinking signature), `input_json_delta` (tool args). `content_block_stop` may include `finish_reason: 'max_tokens'` to trigger auto-continuation.

The consumer processes one tool call per agent step — providers should emit only the first tool_call when a model returns parallel calls.

### Model Data Flow
1. **Model Registry** (`utils/model_registry.ts`) — fetches litellm's `model_prices_and_context_window.json` from GitHub, caches locally with stale-while-revalidate (24h stale / 7d expire / ETag conditional requests). Provides `getModel(id)` for O(1) lookups of pricing, context, and capabilities (vision, tool use, audio, video, reasoning, etc.).
2. **Reasoning Effort** (`utils/reasoning_effort_data.ts`) — centralized profiles for reasoning effort preferences. Different providers use different formats (OpenAI: low/medium/high, Anthropic: token budgets, Gemini 2.5: auto/budget tokens, Ollama: enabled/disabled). `getReasoningEffortPreference(modelId, provider?)` returns the correct dropdown config.
3. **Model Fetchers** (`api/models/*.ts`) — each provider fetches its model list from the API (or uses static data), then enriches with registry data and reasoning effort preferences.

## API Reference

Just use the `local_api` tool to request these APIs.

| Endpoint | Description |
|----------|-------------|
| `llm/enconvo_ai` | Chat using EnconvoAI which provide LLM service, , learn more : [docs](https://www.enconvo.com/cloud-plan). _No params_ |
| `llm/models/1minai` | Fetch and search 1min AI model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/aimagicx` | Fetch and search Aimagicx model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/anthropic` | Fetch and search Anthropic model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/azure` | Fetch and search Azure OpenAI model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/cerebras` | Fetch and search Cerebras model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/dashscope` | Fetch and search DashScope model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/deepseek` | Fetch and search DeepSeek model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/enconvo_ai` | Fetch and search Enconvo AI model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/fetch` | Fetch and search model list from a generic OpenAI-compatible endpoint. _5 params — use `check_local_api_schemas` tool_ |
| `llm/models/fireworks` | Fetch and search Fireworks model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/gemini` | Fetch and search Gemini model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/groq` | Fetch and search Groq model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/minimax` | Fetch and search Minimax model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/mistral` | Fetch and search Mistral model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/mlx` | List MLX-LM models served by the local mlx_manage extension.. _No params_ |
| `llm/models/moonshot` | Fetch and search Moonshot model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/ollama` | Fetch and search Ollama model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/open_ai` | Fetch and search OpenAI-compatible model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/openrouter` | Fetch and search OpenRouter model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/poe` | Fetch and search Poe model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/qwen` | Fetch and search Qwen model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/sambanova` | Fetch and search SambaNova model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/straico` | Fetch and search Straico model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/vercel_ai_gateway` | Fetch and search Vercel AI Gateway model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/x_ai` | Fetch and search xAI model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/xiaomi_mimo` | Fetch and search Xiaomi MiMo model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |
| `llm/models/z_ai` | Fetch and search Z.AI model list. Params: `forceRefresh` (boolean, default: false), `query` (string) |