Launch timeline updated
2mNorthstar moved from Tuesday to Friday.
{ the memory engine }
Memory that gets smarter,
not just bigger.
Turn every interaction into durable, structured context. MatiAI understands what matters, resolves what changed, and recalls what is useful now.
Explore the architecture What changed this week?
RELEVANT CONTEXT
Communication preference
1dShare a concise summary in Slack.
A COMPLETE MEMORY PIPELINE
Less plumbing. Better context.
Everything from ingestion to deletion is one coherent system, with control at every step and no external database to run.
Automatic extraction
Convert conversations, documents, and events into clean facts, preferences, and relationships — no manual labeling required.
Learn moreEvolving memory
Update beliefs as new information arrives. Preserve provenance and resolve contradictions without rewriting history.
Learn moreHybrid retrieval
Combine semantic, keyword, temporal, and graph signals for precise, low-latency context on every call.
Learn moreMemory graph
Connect people, projects, decisions, and events without standing up or maintaining a graph database.
Learn morePortable by default
Export memories as JSON, stream changes, or move between cloud and self-hosted at any time.
Learn moreSafe forgetting
Set retention, decay, and deletion policies by user, namespace, or memory type — down to a single fact.
Learn moreIdentity & consent
Bind memories to real people with explicit scopes, consent flags, and per-user visibility controls.
Learn moreRegional storage
Pin data to US, EU, or APAC regions for residency, latency, and compliance requirements.
Learn moreRealtime sync
Stream every memory change to webhooks, queues, and downstream systems the moment it happens.
Learn moreMEMORY PIPELINE
Six stages, one continuous loop.
Memory is not a single write to a database. It is a living pipeline that ingests, understands, reconciles, stores, recalls, and forgets — on every call.
01
Ingest
Stream conversations, documents, webhooks, and events. MatiAI accepts raw text, structured JSON, and transcripts.
02
Extract
Turn input into typed memories — facts, preferences, events, and relationships — with confidence and provenance.
03
Resolve
Reconcile new memories against existing ones. Detect contradictions, updates, and duplicates automatically.
04
Store
Persist to a tenant-isolated, region-pinned store with full version history and source references.
05
Recall
Retrieve the right context for any prompt with hybrid semantic, keyword, temporal, and graph ranking.
06
Forget
Apply retention, decay, and deletion policies so memory stays sharp, current, and compliant.
MEMORY, NOT STORAGE
Know what changed and why.
MatiAI keeps the meaning, source, confidence, and history behind every fact. When users change their minds, their memory changes with them — and nothing is lost.
- Source-level provenance
- Automatic conflict resolution
- Memory version history
- Configurable confidence thresholds
- Per-user consent & visibility
- Regional data residency
Memory workflowProduction / live pipeline
ActiveConversationWebhook trigger
Extract factsMemory processor
Resolve contextConflict policy
MatiAIStore and recall
ARCHITECTURE
Six layers, zero databases to run.
MatiAI replaces the patchwork of vector stores, embeddings pipelines, and graph databases with a single, layered engine you control through one API.
L1 · Ingestion
Ingestion layer
High-throughput intake from any source, with backpressure and ordering guarantees.
- Webhooks
- Streaming batch API
- Document connectors
- CLI + SDKs
L2 · Extraction
Extraction layer
Model-assisted extraction that produces structured, typed memories with confidence.
- Fact extraction
- Entity linking
- Preference detection
- Event parsing
L3 · Graph
Memory graph
A relationship store that connects people, projects, decisions, and time.
- Entity nodes
- Typed edges
- Temporal edges
- Contradiction tracking
L4 · Retrieval
Retrieval layer
Hybrid ranking that blends signals for precise, low-latency context.
- Semantic search
- Keyword (BM25)
- Temporal recency
- Graph traversal
L5 · Control
Control plane
Namespaces, users, retention, consent, and access — managed through one API.
- Tenant isolation
- Consent scopes
- Retention policies
- Audit logs
L6 · Data
Data plane
Region-pinned, encrypted storage with portable export and streaming sync.
- AES-256 at rest
- Regional residency
- Version history
- Export to JSON
HOW IT COMPARES
MatiAI vs. the usual stack.
The difference between a memory engine and a vector database is everything you do not have to build, operate, or debug.
| Capability | MatiAI | Raw vector DB | RAG from scratch | Chat history |
|---|---|---|---|---|
| Durable, structured memory | ✓ | ✗ | Partial | ✗ |
| Automatic conflict resolution | ✓ | ✗ | Manual | ✗ |
| Provenance & version history | ✓ | Limited | Manual | ✗ |
| Built-in forgetting & decay | ✓ | ✗ | Custom build | ✗ |
| Memory graph (no infra) | ✓ | ✗ | DIY | ✗ |
| Per-user consent controls | ✓ | Partial | DIY | ✗ |
| Typed SDKs + MCP | ✓ | ✗ | ✗ | ✗ |
| Portable export | ✓ | Limited | ✗ | ✗ |
RETENTION & DECAY
Memory that stays sharp.
Storage grows cheap; signal degrades. MatiAI lets you decide what to keep, what to fade, and what to forget — automatically.
Retention windows
Set TTLs by memory type, user, or namespace. Keep decisions forever, fade small talk weekly.
Decay curves
Confidence decays over time unless reinforced, so stale memories naturally lose weight in retrieval.
Conflict policy
When new facts contradict old ones, MatiAI resolves, supersedes, or flags for review — never silently overwrites.
WHAT GETS REMEMBERED
Five memory types, one store.
Every memory is typed on the way in, so retrieval can filter by what kind of context you need — not just what is semantically close.
Preferences
How a user likes things done.
Concise repliesAsync updatesDark mode
People
Who matters and how they relate.
TeammatesStakeholdersContacts
Projects
Active work and its status.
Northstar launchQ3 roadmapMigration
Decisions
What was decided, and why.
Ship FridayUse PostgresPause v2
Events
When things happened.
Standup notesOutagesReleases
{ Start building }
Give your AI a memory it can trust.
Free to start. Open by design. Ready for production.
Get MatiAI