Developer Documentation
Build with SynthCaaS as your memory substrate
SynthCaaS as a Memory Substrate
SynthCaaS provides a persistent memory layer that sits beneath your AI applications. Instead of building bespoke memory systems, you connect to SynthCaaS and get:
- Unified user context — All user data sources normalized into a consistent format
- Semantic memory search — Query memories by meaning, not just keywords
- Memory provenance — Every artifact traces back to its source
- Reinforcement signals — Built-in feedback loops for continuous improvement
Integration Options
MCP Connector
Zero-config integration with any MCP-compatible AI system. Context flows automatically.
REST API
Direct HTTP access for full control. Typed SDKs available for TypeScript, Python, Go.
Webhooks
Real-time event streams for reactive applications. Signed payloads with retry logic.
Dataset Export
Export memories as JSONL or Parquet for fine-tuning, evaluation, or analysis.
MCP Integration
The Model Context Protocol (MCP) provides a standardized way for AI applications to access context. SynthCaaS implements an MCP server that exposes your memory layer to any compatible client.
Add SynthCaaS to your MCP configuration to enable automatic context injection:
Once configured, any MCP-compatible AI will automatically receive relevant user context from SynthCaaS.
REST API Quickstart
Core Endpoints
| Method | Endpoint |
|---|---|
| POST | /v1/connectors/:provider/authorize |
| GET | /v1/context/snapshot |
| POST | /v1/memory/search |
| POST | /v1/feedback |
| POST | /v1/export/jobs |
Fetching User Context
Searching Memory
Posting Feedback
Webhook Events
Subscribe to real-time events to build reactive applications. All webhook payloads are signed with HMAC-SHA256 for security.
Event Types
memory.created— New memory artifact generatedmemory.updated— Existing memory modifiedmemory.deleted— Memory artifact removedconnector.synced— Data source completed syncconnector.error— Data source sync failedexport.completed— Dataset export ready
Dataset Export
Export your memory layer as structured datasets for model fine-tuning, evaluation, or analysis.
Supported Formats
- JSONL — Line-delimited JSON, compatible with most ML frameworks
- Parquet — Columnar format for efficient analytics (coming soon)
Export Options
- Filter by date range, source, or memory type
- Include or exclude embeddings
- Add feedback signals for RLHF training
- Generate eval harness bundles
Ready to Build?
Request early access to get your API keys and start integrating SynthCaaS into your application.
Get Early Access