Feature / Open Source AI Memory Database

Multimodal Memory for AI Agents

Memorose accepts multimodal event streams so agents can build memory from text, image, audio, video, and structured JSON instead of only chat transcripts.

Read the docs Star on GitHub

The problem this solves

Real agent workflows often include screenshots, logs, clips, attachments, and structured traces. A useful memory runtime needs to preserve this context in a durable, searchable form.

How Memorose handles it

The runtime ingests stream-scoped events and consolidates them into memory units that can later be retrieved with semantic and structured controls.

runtime example

curl -X POST "$BASE_URL/v1/users/dylan/streams/support-session/events" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content_type": "json",
    "content": {"ticket": "latency investigation", "priority": "high"}
  }'