Product Surfaces
Aionis exposes product surfaces around one Runtime loop:
observe -> guide -> agent action -> feedback -> measure -> snapshotThe product center is Execution Memory: route-safe, execution-ready context compiled from long-running history. Memory Firewall and Flight Recorder are adjacent surfaces built on the same admission and audit substrate.
The main public surfaces are ordered by product role:
| Surface | Role | Use it when |
|---|---|---|
| Execution Memory | Core Runtime capability | Agents need route-safe context across sessions, roles, compaction, and handoffs. |
| Trace-to-Skill Candidates | Learning surface inside Execution Memory | Verified execution traces can become reviewable skill or procedure candidates without entering the prompt automatically. |
| Memory Firewall | Governance entry point | You already have Mem0, Zep, vector DB, markdown, or another backend and need safe admission before prompt use. |
| Agent Flight Recorder | Audit and replay | You need to know what the Agent could see, what was blocked, and why. |
| Loop Engineering | Loop memory layer | Your Agent already plans, acts, validates, repairs, and needs memory across iterations. |
| Controlled Forgetting | Lifecycle control | You need suppression, archival, rehydration, and deletion without silent memory drift. |
| Admission Dataset Export | Learning flywheel | You want to export admission, usage, outcome, and feedback data for analysis or learned policy training. |
Product Positioning
Execution Memory is the primary Aionis memory system. It records task evidence, adjudicates active state, preserves validation boundaries, and compiles a governed Agent context.
Memory Firewall is the fastest adoption path. It lets teams keep their existing memory backend while using Aionis as the admission layer. That makes Firewall a strong entry point, but not the ceiling of the product.
The other surfaces build on the same substrate: decision traces, feedback attribution, lifecycle state, and auditable context compilation.
Trace-to-Skill Candidates are intentionally nested under Execution Memory. They expose measured positive traces as controlled candidate assets while preserving Aionis’s normal rule: a candidate is not current route state until admission, feedback, and promotion gates justify later use. The first product path is review-first: trace -> feedback attribution -> measure -> candidate -> review -> promotion gate.