The trust layer for autonomous AI
XMDB is a proof of provenance substrate providing auditable, cryptographically verifiable history, long-term memory, retrieval, and coordination across agentic systems.
AI agents lose state in production. Context windows overflow. Memory fragments across tools and runs. Most agents cannot reliably resume work where they left off. In multi-agent systems it is worse. There is no shared source of truth. There is no verifiable record of what each agent knew or did.
XMDB gives agents persistent long-term memory and shared coordination primitives. It includes durable mailboxes, work queues, pub/sub topics, leases, and presence. It is built on an append-only, hash-chained event log. Replay any run, explain any decision, and verify integrity with one click.
- Long-term memory across sessions and restarts
- Multi-agent coordination via mailboxes, queues, topics, and leases
- Append-only event log, hash-linked for tamper evidence
- Browse history and verify any event with one click
AI memory today is a liability.
Agents forget what they learned. State disappears between runs. Multiple agents can't share context without brittle workarounds. And when something goes wrong, there's no record of what happened or why.
Even when assistants remember user preferences, teams still lack shared, policy-controlled, audit-ready memory for autonomous systems.
For teams building production agent systems this is not a minor gap. Lack of auditability blocks production.
Memory as a ledger, not a cache.
XMDB treats agent memory as durable, append-only history instead of a cache that gets overwritten. Vector search is available for retrieval, but the hash-linked event log is the source of truth. Browse the full chain and verify any event's integrity with one click.
Immutable history
Every write is appended, hash-linked, and optionally signed. History can't be rewritten.
Deterministic retrieval
Context packs are attested with pack_hash and signatures. Same query, same results.
Verifiable provenance
Every decision traces back to its inputs, policy context, and retrieval state.
Tracked beliefs
Record what agents concluded, from what evidence, with what confidence. Beliefs can be superseded or invalidated while preserving a complete audit trail.
Proof at every layer
Every step of an agent run is cryptographically verifiable. This includes what it stored, what it retrieved, and what it concluded.
Receipts
Proof of retrieval
Every context pack includes a cryptographic receipt. It proves exactly what the agent saw before making a decision.
Beliefs
Tracked conclusions
Record what agents concluded, why they believed it, and when they changed their mind. Query beliefs at any point in time.
Merkle proofs
Selective verification
Prove any specific event exists without exposing the full chain. Auditors verify independently.
External anchoring
Third-party attestation
Anchor Merkle roots to RFC 3161 timestamp authorities. Cryptographic proof that data existed at a specific time, verified by a trusted third party.
Federation
Multi-agent trust
Share beliefs between agents with cryptographic proof of the entire reasoning chain. Trust policies control what you accept and why.
Audit-ready by default.
XMDB gives your compliance and security teams what they need without bolting it on later.
- Append-only event log with cryptographic integrity
- Explainability mode for any retrieval
- Retention and lifecycle controls
- Fine-grained access control per workspace, scope, and policy
- On-prem deployment or encrypted cloud sync. Your choice.
- Merkle proofs for selective verification
- External timestamp anchoring via RFC 3161
- Belief tracking with full temporal history
- Point-in-time queries that reconstruct the exact state at any moment
Shared memory. Controlled access.
Multiple agents connect to the same memory substrate. Writes are visible to all and retrieval is controlled by policy. No more duct-taping state between systems.
Agents via MCP (examples)
Any MCP-compatible agent can connect.
Systems (examples)
Connect any system via API, SDK, or webhooks.
Logos shown are examples. Product names are trademarks of their respective owners.
Agent Coordination
XMDB turns agent collaboration into a durable, inspectable system.
XMDB provides a shared coordination layer for autonomous systems. Agents communicate through durable mailboxes and topics, coordinate work with leasing, and record every step as ordered history that can be streamed, audited, and replayed.
Mailboxes & topics
Agent-to-agent messaging and topic pub/sub with deterministic cursors, ACKs, and SSE streams.
Work queue leasing
Claim/renew/release semantics with deterministic work state.
Threads, runs & steps
Structured collaboration ledger with idempotent execution.
Presence & capabilities
Heartbeat-based presence and discoverable agent metadata.
Webhooks
Signed, retrying webhook delivery with delivery history.
Governance & export
Audit-friendly exports and retention controls.
XMDB vs. the alternatives
Vector databases are search indexes. XMDB is the source of truth.
| Capability | Vector DB / RAG | Memory platforms | XMDB |
|---|---|---|---|
| Semantic recall | Yes | Yes | Yes |
| Canonical system history | Partial | Partial | Yes |
| Tamper-evident log | No | No | Yes |
| Deterministic retrieval | No | No | Yes |
| Audit trail | Partial | Partial | Yes |
| Multi-agent coordination | No | Partial | Yes |
| On-prem option | Partial | Partial | Yes |
| Verifiable provenance | Partial | Partial | Yes |
| Interactive integrity verification | No | No | Yes |
| Belief tracking | No | No | Yes |
| Merkle proofs | No | No | Yes |
| External timestamp anchoring | No | No | Yes |
| Point-in-time queries | No | Partial | Yes |
| Multi-agent federation with proofs | No | No | Yes |
| Fine-grained access control | Partial | Partial | Partial |
Yes = first-class primitive · Partial = possible with glue · No = not supported
How it works
XMDB records system memory as a persistent history rather than constantly overwritten state. The event log is the source of truth; derived indexes enable fast retrieval.
Works with your stack.
Native MCP support for Claude, Cursor, ChatGPT, Gemini, and more. Connect any system via API, SDK, CLI, or webhooks.
Ready to ship agents you can trust?
Talk to us about your use case. We'll show you how XMDB fits your memory, coordination, and compliance requirements.