Context for AI, simplified
CLAIV Memory returns context (system_context + memory_blocks) for your LLM. Your LLM generates the final reply. Tenant is inferred from your API key.
Ingest Events
Send messages, tool calls, and app events via a simple POST. Each event is stored and queued for enrichment. Returns event_id and deduped status.
POST /v1/ingest
{
"user_id": "user-123",
"type": "message",
"content": "Prefers dark mode"
}
=> { "event_id": "evt_abc", "deduped": false }Async Enrichment
A background worker enriches events asynchronously (1-5 seconds). It extracts facts, episodes, and generates embeddings. Recall may return empty until enrichment completes.
// Worker processes in background:
// 1. Extract facts & episodes
// 2. Generate embeddings
// 3. Deduplicate against existing data
// 4. Update memory graph
// Typically completes in 1-5 secondsRecall Context
Retrieve relevant context within a token budget (200-8000). CLAIV returns system_context + memory_blocks. Your LLM generates the final reply.
POST /v1/recall
{
"user_id": "user-123",
"task": "Get UI preferences",
"token_budget": 2000
}
=> {
"system_context": "User prefers dark mode...",
"memory_blocks": [...],
"citations": ["evt_abc"],
"token_estimate": 42
}Forget with Receipts
Delete derived data (facts, episodes, embeddings) with cryptographic receipts for GDPR/CCPA compliance. Original events preserved for audit.
POST /v1/forget
{
"user_id": "user-123"
}
=> {
"receipt_id": "rcpt_xyz789",
"deleted_counts": {
"facts": 3, "episodes": 1, "embeddings": 12
}
}Built for production AI
Everything you need to add persistent context to your AI applications
Three Event Types
message, tool_call, app_event. Ingest every interaction your AI has.
Token Budget Control
Specify exactly how many tokens of context to return (200-8000). No more context overflow.
Compliance Built-in
Every forget returns receipt_id + deleted_counts. Audit-ready for GDPR/CCPA.
Simple REST API
Three endpoints: ingest, recall, forget. Tenant inferred from API key.
Ready to get started?
Join the beta and be among the first to build with persistent context.