SPM Documentation
SPM (StellarPath Memory) provides memory and context infrastructure for AI agents through a hosted provider proxy and governed long-term memory. Configure any OpenAI- or Anthropic-compatible client to use the SPM endpoint while retaining your existing LLM provider account. Each request receives evidence-gated recall, deterministic context compression, and a signed audit trail.
SPM is a hosted service. It does not currently offer an appliance or self-deployment: integrate through the provider proxy or MCP server while we operate the memory plane.
Two doors in
| Door | Endpoint | For |
|---|---|---|
| Provider proxy | https://api.spmos.ai/v1 |
Any OpenAI/Anthropic-compatible app or agent |
| MCP server | https://api.spmos.ai/mcp |
Codex, Claude Code, and other MCP clients |
What SPM does inside a request
- Authenticates your API key and resolves your tenant, policy, and provider.
- Recalls relevant memory through an evidence gate. If nothing qualifies, the gate fails closed instead of guessing.
- Compresses the assembled context deterministically. The memory path does not call an LLM, so memory growth does not increase inference costs.
- Forwards to your provider in its native dialect and streams the response back.
- Ingests the exchange asynchronously through idempotent jobs, retries, and dead letters, then writes a signed request receipt.
Where to go next
- Quickstart — first call in five minutes
- Authentication & API keys — keys, scopes, and token hygiene
- Bring your own provider — wire DeepSeek, OpenAI, Azure, Anthropic, Gemini
- Memory, evidence & deletion — the gate, provenance, and signed purge receipts
- Benchmarks — the scorecard, with footnotes