Agnes AI Models is the official gateway and model catalog for Agnes AI, a company that trains its own text, image, video, and reasoning models. Developers reach the models through an OpenAI-compatible API at https://apihub.agnes-ai.com/v1 with bearer-token authentication, so existing OpenAI client libraries work by changing the base URL. The catalog positions its chat models for agentic tools such as OpenClaw and Hermes.
The repository holds the public documentation set: a model catalog, changelog, troubleshooting and error-code guides, an FAQ, Token Plan details, and minimal curl, Python, and Node.js examples covering streaming chat, image generation, asynchronous video generation with video_id polling, and a tool-calling agent workflow. Reference limits are dated and may change; production values should be confirmed in the platform console.
| Model | Type | Endpoint | Capabilities |
|---|---|---|---|
| agnes-2.5-flash | Text and vision-language | /v1/chat/completions | Coding, agent workflows, tool calling, multi-turn dialogue, reasoning, image understanding; 512K context, 65.5K max output |
| agnes-2.0-flash | Text and vision-language | /v1/chat/completions | Reasoning, coding, tool calling, streaming, image understanding; 256K context, 64K max output |
| agnes-image-2.0-flash | Image generation and editing | /v1/images/generations | Text-to-image, image-to-image, URL or Base64 output |
| agnes-image-2.1-flash | Image generation and editing | /v1/images/generations | Image editing, flexible sizes, URL or Base64 output |
| agnes-video-v2.0 | Video generation | /v1/videos | Text-to-video, image-to-video, multi-image video, keyframe animation, async task API |
| User plan | Text RPM | Image RPM | Video |
|---|---|---|---|
| Free / default | 20 | Resolution-specific limits | 1 RPM |
| Enterprise | 40 | Higher resolution-specific limits | 2 RPM |
| Token Plan | 1,000 | Higher 1K and 2K limits | 5 RPM; 500 seconds per day |
Features
- OpenAI-compatible gateway: the same base URL, request format, streaming, tool calling, and image URL input across chat models
- Chat and vision models: agnes-2.5-flash and agnes-2.0-flash with thinking mode, streaming, tool calling, and image inputs
- Image generation: text-to-image and image-to-image with URL or Base64 output
- Video generation: asynchronous task creation on /v1/videos, polled by video_id
- Subscription quotas: Starter, Plus, and Pro tiers with per-5-hour and weekly request caps, 4,000 images per day, and 500 seconds of video per day
- Regional sites: an international site and a China site with the same developer docs and API platform
- Error guidance: documented handling for 400, 401, 429, and 5xx responses, including backoff recommendations
