All docs
Docs
Getting Started

Start here

A short mental model for how Synapsor differs from a normal database, vector database, or agent framework.

Overview

A normal database stores and queries data. A vector database retrieves relevant chunks. An agent framework calls tools. Synapsor adds the durable trust boundary for agent data access: trusted session context, reviewed capabilities, durable run graphs, evidence, branches, write proposals, external action intents, approvals, and replay.

Hosted V2 is controlled-beta, single-node software. Free remains invite-approved; Builder starts with a 14-day no-card trial for approved beta users. It is useful for evaluation and prototypes where auditability matters, but it does not promise HA, SLA, compliance, serverless storage separation, or horizontal scaling.

The first workflow to learn is simple: bind trusted session values, create an agent context, create an agent workflow, invoke reviewed capabilities inside a run, inspect evidence, then use proposal mode for risky writes.

Agent execution lifecycle

Synapsor does not run autonomous agent loops. It records and governs each capability invocation that your app or agent framework calls.

1
User request

Your app receives the request and decides which reviewed capability to call.

2
Session bindings

Tenant, principal, and current object ids are bound server-side, outside prompt control.

3
Agent context

Synapsor looks up rows, runs hybrid search, recalls governed memory, and prepares output slots.

4
Capability

A read-only or proposal capability executes under RBAC, quota, and evidence policy.

5
Evidence/replay

The invocation returns handles that can be inspected, audited, and replayed later.

Where Synapsor fits

Use Synapsor for the parts of an agent app that must be durable, governed, and replayable.

Normal DB
  • Tables, indexes, transactions
  • App owns agent context glue
  • Audit must be reconstructed
Vector DB
  • Chunk retrieval and similarity search
  • Weak write lifecycle
  • Limited policy/replay state
Agent framework
  • Conversation loop and tool calling
  • Runtime orchestration
  • Not the source of durable DB policy
Synapsor
  • Data plus agent execution state
  • Evidence, proposals, branches, settlement
  • Replayable capability invocations

Developer notes

  • Use hidden session bindings for trusted values.
  • Use read-only capabilities for answers and proposal capabilities for writes.
  • Store evidence lookup records with user-visible decisions.
  • Start with Free Controlled Beta, then request a Builder Trial invite when you need higher beta limits.