First Agent Loop

Aionis becomes useful when your Agent loop reports what happened, asks for governed context, and sends outcome attribution back.


observe -> guide -> agent action -> feedback -> measure -> snapshot

Minimal Shape


import {
  agentPromptFromGuide,
  feedbackFromGuide,
} from "@aionis/sdk";
 
const guide = await aionis.guide({
  query_text: "Continue from the latest accepted route.",
  context_mode: "compact_agent",
  task_context_profile: "coding_verifier",
});
 
const result = await agent.run(agentPromptFromGuide(guide));
const usedMemoryIds = result.usedMemoryIds ?? [];
 
if (usedMemoryIds.length > 0) {
  await aionis.feedback(feedbackFromGuide({
    guide,
    run_id: result.runId,
    outcome: result.ok ? "positive" : "negative",
    reason: result.summary,
    used_memory_ids: usedMemoryIds,
    used_surface: "use_now",
  }));
}

What This Proves

Step	Why it matters
`observe`	Records evidence instead of only storing a text note.
`guide`	Compiles current, inspect, suppressed, and rehydrate surfaces.
Agent action	Lets your own Agent host decide and execute.
`feedback`	Attributes outcome only to memory exposed by that guide.
`measure`	Reports whether memory helped or created risk.
`snapshot`	Produces replayable operator context.

Choose A Context Profile

Your host can set task_context_profile when it knows the task posture. The profile changes prompt rendering only; Aionis still uses the same governed memory surfaces.

Workload	Profile
Coding with checks or validators	`coding_verifier`
Long-memory QA	`long_qa`
Multi-agent handoff	`multi_agent_handoff`
Plan/validate/repair loop	`loop_engineering`
File or document identity work	`document_integrity`

See Task Context Profiles.

Optional Local Verification


npm run -s runtime:quickstart:sdk

Use this only to verify the installed Runtime path. For production integration, wire the same observe -> guide -> feedback -> measure -> snapshot loop into your own Agent host, then choose the host path in Choose an Integration.