Raw recall117k
117k full-history prompt
Failed branch appears relevant
Stale target competes with current route
Agent must rediscover what happened
Aionis Runtime v0.2
Execution memory runtime for AI agents
Aionis turns Agent execution history into governed state: what to use now, what to inspect, what to block, what to rehydrate, and what actually helped. Use it through SDK/API, CLI, or MCP to keep Agents working across sessions, compaction, handoffs, and model switches.
RAW STATE
SAFE CONTEXT
PlansDecisionsFailed branchesHandoffsExternal memory
State - Context - Receipt
use_nowinspect_before_usedo_not_userehydrateReceipts
Before / After
From noisy recall to executable context.
Aionis compiles long execution history into a smaller action contract: accepted state, blocked branches, rehydrate pointers, and an audit receipt.
Aionis context28k
use_nowaccepted active route
do_not_usefailed branch kept as counter-evidence
rehydrateraw payload available on demand
receiptmemory influence remains auditable
Record execution state
Capture plans, decisions, branches, tool outcomes, verifier results, and handoffs while the Agent works.
Compile the next context
Route memory into use_now, inspect_before_use, do_not_use, and rehydrate before it reaches the prompt.
Attribute outcomes
Connect surfaced memories, suppressed memories, guide traces, feedback, and measurement reports.
Why it exists
Retrieval finds similar text. Agents need execution state.
Raw memory recall is not enough for long-running Agents.
Aionis turns history into an action contract before the next prompt is compiled, so the Agent can continue from the route that actually survived validation.
Long tasks lose the route
After compaction or a new session, the Agent needs the accepted plan, active path, and current constraints.
Relevant memories can mislead
A failed branch can look semantically relevant. Aionis keeps it visible as evidence without turning it into instruction.
Teams need an audit trail
When an Agent acts, operators need to know which memory influenced it, which was blocked, and what outcome followed.
Admission Classifier
Judge memory before it acts.
Aionis does more than retrieve relevant memory. It classifies whether each candidate is usable for the current execution state, then routes it into an explicit action surface.
use_now
Accepted route
The compact state the next agent can use immediately.
inspect_before_use
Needs review
Unknown, contested, or weak evidence stays visible but gated.
do_not_use
Blocked memory
Failed branches and stale context are retained as counter-evidence.
rehydrate
Bring back detail
Pointer-only history can be restored without flooding the prompt.
Runtime
Give every agent loop a durable execution state.
Aionis plugs into plan, act, validate, and revise loops. It records what happened, compiles the next context, attributes outcomes, and measures whether memory helped.
Observe
Record plans, tool calls, file targets, validation results, decisions, and branch changes while the agent works.
Adjudicate
Turn raw history into use_now, inspect_before_use, do_not_use, and rehydrate-ready execution state.
Resume
Compile a compact context packet for the next model, session, teammate, or coding agent.
Products
Three core products, one runtime layer.
Admitted memory, replayable influence, and compact state-preserving context.
Primary surface
Memory Firewall
Admission Policy judges whether memory is usable for the current execution state before it enters the Agent prompt.
Agent Flight Recorder
Replay what memory the Agent could see, what was blocked, why it was surfaced, and what outcome followed.
State-Preserving Context Compression
Compress long execution history while preserving current state, blocked branches, procedure memory, and rehydrate pointers.
Use cases
Where teams apply it first.
Coding Agent Execution Memory
Keep coding tasks moving after session breaks, compaction, and model switches by resuming from accepted execution state.
Loop Engineering Memory Layer
Persist the plan, act, validate, repair, and stop conditions that make agent loops improve instead of restart from scratch.
Multi-Agent Handoff
Give planner, worker, verifier, and reviewer agents the same governed state without copying raw transcripts between roles.
Install
Download Runtime. Add SDK/API.
Start with one command, then connect Aionis to your Agent loop. MCP is available for coding tools, but the core path is the Runtime API and SDK.
- Use the SDK/API in any custom Agent loop
- Continue long-running work after context compaction
- Carry execution state across planner, worker, verifier, and reviewer handoffs
- Govern memory from Aionis, Mem0, Zep, Supermemory, Chroma, pgvector, logs, and internal stores
aionis-runtime
$ npx @aionis/create$ npm i @aionis/sdk$ npx @aionis/mcp@latest --scope my-projectBenchmarks
Measured, not guessed.
Aionis reports context cost, completion, admission risk, and memory influence with repeatable evaluation paths.
28k / 117k776 rows70 to 05 arms
Context pressure
Buried-history runs compare compact governed context against full-history prompts when continuation tasks get noisy.
Open reportAdmission dataset
Guide receipts become labeled rows for policy comparison, holdout checks, and real LLM reruns.
Open reportClosed-loop replay
Prior-aware negative direct-use cases are cleared in replay when closed-loop state is admitted through the policy gate.
Open reportAgent E2E
Aionis is compared with no memory, full history, BM25, and retrieval memory on continuation tasks with planted traps.
Open reportWhy Aionis
What changes after you connect it.
Aionis turns memory from raw retrieval into an execution-state contract the Agent can act on.
Agents resume from accepted state
The next run gets the current plan, active route, handoff notes, acceptance checks, and rehydrate pointers instead of a raw transcript.
Memory is admitted before prompt use
Relevant candidates are routed into use_now, inspect_before_use, do_not_use, or rehydrate so stale and failed history stays as counter-evidence.
Every memory influence is traceable
Receipts connect guide traces, surfaced memories, suppressed memories, feedback outcomes, and measurement reports.
FAQ
Quick answers.
Install the Runtime, connect the SDK/API, and let Aionis compile the next Agent context.
What does Aionis do?
It records what an Agent planned, tried, verified, failed, and handed off, then compiles the next prompt with the state that should influence action.
How do I add it to my Agent?
Install the Runtime, use the SDK or HTTP API from your Agent loop, call guide before the next action, and send feedback after the outcome.
Where does MCP fit?
MCP is the fastest zero-code bridge for tools like Claude Code, Cursor, and Zcode. It is one access path, while SDK/API is the main product integration path.
Why use it with existing memory systems?
Retrieval finds candidates. Aionis decides whether each candidate is usable now, inspect-only, blocked, or should be rehydrated with more evidence.
Give your coding agents executable memory.
Install the Runtime, wire the SDK/API, and bring governed execution memory into your Agent workflow.