letta-provider-umans

Umans.ai provider for Letta Code. Adds subscription-based access to coding-optimized models including Kimi K2.7, GLM 5.2, and more.

Install

Ask your Letta Code agent to install this mod:

Install the Umans provider mod from npm:letta-provider-umans

If you plan to use GLM 5.1 or GLM 5.2, ask your agent to also install the image-understanding mod for vision handoff:

Install the Umans provider mod from npm:letta-provider-umans and also install the npm:@letta-ai/image-understanding mod for vision handoff

Or install manually:

letta install npm:letta-provider-umans

Then reload:

/reload

Connect

Get an API key from app.umans.ai/billing → Dashboard → API Keys, then either:

/connect umans

Or set an environment variable:

export UMANS_API_KEY="uk-your-key-here"

Models

Discovered live from the Umans gateway. New models appear automatically after /reload — no mod update needed.

| Model | Context | Vision | Reasoning | | --- | --- | --- | --- | | umans-coder | 262K | native | yes (always on) | | umans-kimi-k2.6 | 262K | native | yes (can disable) | | umans-kimi-k2.7 | 262K | native | yes (always on) | | umans-glm-5.1 | 202K | via handoff | yes (can disable) | | umans-glm-5.2 | 405K | via handoff | yes (can disable) | | umans-flash | 262K | native | yes (can disable) | | umans-qwen3.6-35b-a3b | 262K | native | yes (can disable) |

Select with /model umans/umans-glm-5.2.

Vision handoff for GLM models

GLM 5.1 and GLM 5.2 don't have native vision. Install the image-understanding mod to bridge images to text:

letta install npm:@letta-ai/image-understanding

The provider uses the OpenAI-compatible endpoint, so the same API key from /connect umans works. Point the image-understanding mod at Umans and pick a native-vision model:

export IMAGE_UNDERSTANDING_PROVIDER=openai-compatible
export IMAGE_UNDERSTANDING_API_KEY="$UMANS_API_KEY"
export IMAGE_UNDERSTANDING_BASE_URL=https://api.code.umans.ai/v1
export IMAGE_UNDERSTANDING_MODEL=umans-kimi-k2.7

To use a different vision model, e.g. umans-flash:

export IMAGE_UNDERSTANDING_MODEL=umans-flash

Enable auto-captioning so images are described before the GLM model sees them:

export IMAGE_UNDERSTANDING_AUTO_CAPTION=1
export IMAGE_UNDERSTANDING_AUTO_MODE=describe

Reload and verify:

/reload
/image-understanding-status

License

MIT