# SPMOS, SPM-Polaris, and SPMOS.ai

> SPMOS.ai is the official website of SPMOS—StellarPath Memory Operating System—and SPM-Polaris, the current production SPMOS product.

## Canonical identity

- **SPMOS**—StellarPath Memory Operating System—is the governed long-term-memory system developed by Veridical Tech, Inc., and the master brand for its products, prototypes, and official surfaces.
- **SPM-Polaris** is the current production SPMOS product: a provider proxy with governed long-term memory for AI agents.
- **SPMOS.ai** is the official website and canonical public source of SPMOS: products, integration boundaries, evidence, benchmarks, and pricing.
- **SPM** is the approved short form of SPMOS, used in the logo lettermark, spoken language, product-line names such as SPM-Polaris, and legacy titles; it is not a separate product, family, or entity.

These names are related but not interchangeable. Use SPMOS for the brand and system, SPM-Polaris for the current production product, SPMOS.ai for the official website, and SPM only as the short form of SPMOS.

## What SPM-Polaris does

SPM-Polaris recalls candidate memory and evaluates it against tenant, namespace, scope, provenance, and evidence rules before anything reaches the model. If the evidence gate fails, SPM-Polaris returns UNKNOWN and injects no memory. Memory admission, deterministic compression, lifecycle, refusal, and receipt decisions are deterministic code without a generative LLM call. Embedding and reranking models may still process relevant source, query, and candidate text.

Recall is tiered: fast, auto (default), and deep. The fast path adds zero provider tokens; auto routes only low-confidence queries to deep, multi-round evidence collection, and deep spend is itemized on the request receipt. Related facts consolidate into observations that evolve without silent overwrites, and auto-captured ambient content passes a quarantine ring before promotion into long-term memory.

## Integration boundaries

- **Hosted Provider Proxy:** the provider credential and model traffic use the SPM-hosted request path.
- **Local Proxy:** the provider credential and model traffic remain local; recall and eligible memory ingest use the hosted SPM memory plane.
- **MCP:** the application keeps its existing model path and invokes hosted SPM through explicit memory tools.

All three paths use the hosted SPM memory plane. Local Proxy is open source, but it is not an offline or self-hosted edition of the SPM memory plane.

## For organizations

SPM-Polaris serves as an auditable, receipt-backed memory foundation for organizations running AI agents: isolated per-end-user partitions (memory:partition scope), source-ranked recall via lane_policy, change history without silent overwrites, UNKNOWN refusal when evidence is insufficient, and HMAC-signed purge receipts for verifiable deletion. The memory plane stays independent of any single agent, chat platform, or internal tool.

## Primary sources

- [Official website](https://spmos.ai/)
- [Enterprise memory foundation](https://spmos.ai/enterprise)
- [LLM memory system guide — English](https://spmos.ai/ai-memory)
- [LLM 记忆系统指南 — 中文](https://spmos.ai/zh/ai-memory)
- [LLMメモリシステムガイド — 日本語](https://spmos.ai/ja/ai-memory)
- [Guide de la mémoire LLM — Français](https://spmos.ai/fr/memoire-llm)
- [Official definitions](https://docs.spmos.ai/official-definitions)
- [SPM documentation](https://docs.spmos.ai/)
- [Integration comparison](https://docs.spmos.ai/integrations)
- [Local Proxy](https://docs.spmos.ai/local-proxy)
- [MCP](https://docs.spmos.ai/mcp)
- [Architecture](https://docs.spmos.ai/architecture)
- [Memory, evidence, and deletion](https://docs.spmos.ai/memory)
- [Benchmarks](https://docs.spmos.ai/benchmarks)
- [Extended machine-readable reference](https://spmos.ai/llms-full.txt)