Memory & Context Assembly

Omni agents maintain persistent, cross-session memory. Rather than relying on stateless prompt engineering, Omni indexes memories and dynamically injects relevant context into active agent prompts.

Memory Lifecycle

Memory Operations

1. Store a Fact

Persist discrete facts, user preferences, or knowledge artifacts:

curl -sS -X POST "https://edge.omnistatic.com/v1/memories" \
  -H "Authorization: Bearer $OMNI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "User prefers concise Python code examples without docstrings."
  }'

2. Search Memories

Perform vector similarity lookups directly:

curl -sS -X POST "https://edge.omnistatic.com/v1/memories/search" \
  -H "Authorization: Bearer $OMNI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "coding style preferences",
    "limit": 3
  }'

3. Assembled Context Builder

Inspect the exact context bundle that Omni would assemble for a given query:

curl -sS -X POST "https://edge.omnistatic.com/v1/context/build" \
  -H "Authorization: Bearer $OMNI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Python formatting",
    "workspace_id": "ws_123"
  }'

Automatic Context Injection (Shimmer)

When calling /v1/chat/completions, Omni automatically performs an index-grounded lookup before forwarding to the model:

• X-Omni-Shimmer header: Reports injection status (injected, skipped, timeout, or no-context).
• X-Omni-Memory-Count header: Number of semantic memories injected.
• X-Omni-Notes-Count header: Number of workspace notes injected.