For organizations
Applications change.
Governed memory stays.
SPM-Polaris is an auditable, receipt-backed memory foundation: governed long-term memory for your AI agents that stays independent of any single agent, platform, or model — so your organization can change the tools on top without losing what it knows.
Definition
What an enterprise memory foundation is
A memory foundation is the governed long-term memory layer underneath your applications. Agents, chat platforms, and internal tools come and go; what the organization has learned should not leave with them. SPM-Polaris runs that layer as a hosted memory plane: every write, recall, refusal, and deletion passes the same gates and leaves the same evidence, no matter which application produced it.
Not a data warehouse
Transactional records, inventory, and analytics stay in your ERP and databases. The foundation holds governed knowledge and decisions, not your ledgers.
Not an agent builder
Your agents and collaboration tools remain yours — any framework, any vendor. The foundation is the memory they share.
Not a foundation model
Your models stay on your own provider accounts, billed by your provider directly. SPM never resells or marks up inference.
Guarantees
What the foundation guarantees
Each guarantee is a behavior of the production system, documented in the primary sources at the end of this page.
Isolated memory per end user
One API key with the memory:partition scope serves many end users; each memory space is fenced across writes, recall, evidence reads, and read tokens.
Official outranks overheard
The lane_policy parameter separates deliberate saves from passive captures: explicit_first (default), blended, or observed_only. Accepted today; server-side lane ordering activates after a paired evaluation gate.
Change recorded, not overwritten
When a preference, configuration, or decision changes, SPM records the current state and keeps the earlier state as history — nothing is silently overwritten.
UNKNOWN instead of invention
When memory cannot support an answer, the evidence gate returns UNKNOWN with a gate reason and injects no memory.
Auditable operations
Requests carry signed receipts; recall reports its evidence, and deep evidence review reports rounds, selector tokens, and latency on every response.
Erasure you can present
Each successful purge returns an HMAC-SHA256 signed receipt stored with the purge record. Deleted identifiers are never reused within the same memory generation — exactly what GDPR-style erasure requests demand.
Scenarios
Where the foundation plugs in
One memory plane; many surfaces. These are scenario framings, not measured outcomes — each must be validated on your own workload.
Support copilots
Each customer's device history, configuration, and prior tickets live in an isolated partition, so support agents never re-ask and answers arrive with evidence attached. Off-policy questions get UNKNOWN, not an invention.
Community & market intelligence
Agents extract recurring issues and changes from community channels into shared memory once, then recall and redistribute them — with what-was-already-reported remembered instead of re-sent.
Operations knowledge
Schedules, logistics details, and project decisions change often; the foundation records the current state and keeps the previous one as history, so "what was it before" always has an answer.
Cross-platform agents
Any MCP-capable agent — or the compatible APIs — can read and write the same governed memory, whichever collaboration platform, model provider, or internal system it runs on.
Boundaries
Honest boundaries
Read these before you evaluate us.
Scope and custody, in plain terms
- The memory plane is hosted; credential custody is your choice across Hosted Provider Proxy, Local Proxy, and MCP.
- Data-residency and deployment questions belong in a conversation, not a landing page — [email protected].
- Performance numbers on this site are bounded first-party observations with named scopes and dates — not guarantees for your workload.
- Source-ranked lane ordering is rolling out behind an evaluation gate; the lane_policy parameter is accepted today.
Evaluation
How to evaluate it
Three steps, no commitment.
01
Self-serve in the console
The free plan covers real usage at small scale; recall depth, capture switches, compression, and budgets are adjustable in Settings.
02
Verify against the docs
Integration boundaries, architecture, benchmarks, and security pages are the canonical references — every claim on this page traces there.
03
Talk to us
Enterprise deployments, partition design, and data-residency questions: [email protected].
Verify the details
Primary sources
Use the matching official page when citing identity, architecture, integrations, evidence, or limitations.
Your tools will change.
Make sure your memory doesn't have to.
Start with one workflow — a support queue, a knowledge silo, a cross-platform agent — and let the evidence decide.