Ember — thermal embedding

Ember is a thermal sense pack on fusion-embedding-2. Same frozen base, one extra modality. It attaches a small thermal adapter pack to the FE2 decoder and encodes thermal-infrared images and their captions into a shared, L2-normalized space (the raw LLM hidden dim). Thermal encodes run with the pack gate open; captions run with the gate closed, exactly as trained, so the base stays untouched for everything else.

What you get

• Thermal image embeddings from EximiusLabs/fusion-embedding-2-ember.
• Caption embeddings in the same space, so text queries retrieve thermal frames and back.
• OpenAI-shaped responses.

Call examples

Embed a thermal image (http url, data URI, or raw base64):

{ "input": { "thermal": "https://example.com/frame.png" } }

Embed a caption (a single string, or a list):

{ "input": { "text": "a person crossing a dark road" } }

Response:

{
  "object": "list",
  "model": "EximiusLabs/fusion-embedding-2-ember",
  "dim": 2048,
  "data": [ { "object": "embedding", "index": 0, "embedding": [ 0.01, -0.02, "..." ] } ]
}

Cosine similarity between a thermal vector and a caption vector scores the match; both are already L2-normalized, so a dot product is the cosine.

Engram

This pack is one of the modalities Engram searches. Engram is the open cross-modal memory layer for physical AI: it indexes a robot's video, audio, and motion into one embedding space and answers questions about it in plain language, including temporal reasoning that retrieval alone cannot.

pip install engram-robomem

Repo: https://github.com/Eximius-Labs/engram | PyPI: https://pypi.org/project/engram-robomem | Playground: https://www.eximiuslabs.com/playground