R
LoginSign Up
Features→Architecture→Documentation→Pricing→
LoginSign Up Free

Stopbuildinguser_memoriesuser_memoriestables.
GiveyourAIagents
continuousmemory.

Self-evolving long-term memory for LLMs and autonomous fleets
with sub-15ms pgvector retrieval.

A clearer way to add memory

WhyRemora

Remora helps your AI remember what matters, find it quickly, and stay responsive.

AI MEMORYALWAYS READY
Your AI remembers
Useful context across every conversation.
FAST RECALLQuick responses
Under 15 ms
Finds the right context
searching your memory
3 relevant memories
Your question
What your user needs
PreferencesBest match
Your roleStrong match
Past contextRelated
CLEAN CONTEXT
Clean context
Similar memories merged
Similarity checkLikely the same detail
0.93
Prefers TypeScript
Uses Next.js
Merged into one useful memory
FLEXIBLE SETUPWorks with your stack
Fits your model
Works with your AI model
Easy to adapt · quick to search
How Remora finds a memory
Explore the path
L2L1L0
Memory path step 4Following related context
Your memories stay together
Works with popular AI models and databases
BACKGROUND WORKNo interruption
No slowdown
Works in the background
While you chat:KEEP TALKING
tok
tok
tok
Your response stays uninterrupted
Developer Experience
Small surface area. Full memory lifecycle.

Add memory without rebuilding your stack.

Write a meaningful fact, retrieve ranked context before a response, and let Remora handle extraction, deduplication, decay, and history behind the API.

Read the integration guideTypeScript · Python
import { RemoraClient } from "@ai-remora/sdk";

const remora = new RemoraClient({ apiKey });

await remora.add({ userId, content });
const context = await remora.context({ userId, query });
context returned as <user_context>
Cognitive capabilities

Memory with structure, not just storage.

Remora turns conversational signals into typed, ranked, and auditable memory. Each capability exists to keep context useful as an application grows.

memory.type

Typed memory

Facts, preferences, profiles, project context, instructions, goals, and decisions are stored as distinct records—not undifferentiated chat history.

FACTPREFERENCEPROFILEPROJECT_CONTEXTGOALDECISION
06
core types
01
shared record
∞
across sessions
confidence

Deduplication

Repeated statements reinforce an existing memory instead of creating another row.

0.93
similarity thresholdconfidence +0.05
recency

Time-aware ranking

Relevance is balanced with importance, confidence, and the age of a memory before it reaches the prompt.

newer contextolder context
single substrate

One record, one source of truth

Vectors, metadata, and version history stay together in PostgreSQL. No second system to synchronize or reconcile.

memory → content
memory → embedding
memory → versions[]
ingestion

Async by default

Conversation ingestion returns 202 Accepted. Extraction and embedding continue off the request path.

worker queue active
scope

Tenant-aware by design

Every read and write is scoped to the organization, project, and external user.

org_idproject_iduser_id
Deployment options

One memory layer, from prototype to production.

Start with the same API and move to stronger isolation, higher limits, and operational support when your application requires it.

Developer

For evaluation

Build and test memory-backed features without changing your application architecture.

$0 / forever
  • Up to 100,000 memories
  • TypeScript and Python SDKs
  • Search Playground access
  • Community support
Create an account

Production

For live applications

Keep ingestion off the request path and serve context from a managed, low-latency memory layer.

$49 / month
  • Up to 2,000,000 memories
  • Asynchronous conversation ingestion
  • Redis context caching
  • TTL and lifecycle controls
Start with Production

Enterprise

For governed deployments

Apply organization-level controls across projects, users, data retention, and access policies.

Custom
  • Multiple projects and namespaces
  • Dedicated database isolation
  • Tenant-scoped access and key management
  • Purge and export workflows
Talk to Remora
All plans use the same memory model, retrieval API, and prompt-context contract.