SPM-Polaris powers the production SPMOS service: a provider proxy and governed long-term memory layer for AI agents. Recalled candidates are admitted only when their evidence and provenance pass the required gates. If none qualify, SPM returns an explicit UNKNOWN result and injects no memory. Hosted proxy requests and state-changing memory operations produce auditable receipts.
Hi Product Hunt! I'm Yang, building SPMOS at Veridical Tech. SPM-Polaris started from a simple frustration: AI agents with long-term memory quietly poison their own context — anything that looks relevant gets admitted, and nobody can audit why. Our answer is governed memory: candidate memory is evidence-gated before admission, and when nothing qualifies, the system returns UNKNOWN and injects nothing. Every request emits an auditable receipt, including token counts for elided tool output. We'd love feedback from anyone running agents in production — especially on where memory governance breaks down for you.