Examples & Proof
Each result below is generated by a real local Runtime path in this repository. These examples are not benchmark leaderboards. They are contract proofs: they show the product loop running end to end and returning bounded artifacts.
Runnable Examples
For a 3-5 minute external walkthrough, start with the Claude Code Demo.
| Example | Run command | What it proves |
|---|---|---|
| SDK quickstart | npm run -s runtime:quickstart:sdk | SDK can complete remember -> guide -> feedback -> measure -> snapshot. |
| HTTP quickstart | npm run -s runtime:quickstart:http | Raw product facade works without SDK helpers. |
| Multi-agent quickstart | npm run -s runtime:quickstart:multi-agent | Planner, worker, verifier, and reviewer can share execution memory. |
| Memory Firewall quickstart | npm run -s runtime:quickstart:memory-firewall | External candidates can be governed before prompt use. |
| Memory Firewall A/B demo | npm run -s runtime:e2e:memory-firewall-ab | Unsafe external candidates are blocked from direct-use context. |
| Claude Code MCP demo | npm run -s runtime:quickstart:claude-code-mcp | The published MCP bridge can record execution evidence, compile context, and replay a decision for Claude Code / Cursor style clients. |
| Flight Recorder incident demo | npm run -s runtime:e2e:flight-recorder-incident | Aionis can replay what the Agent could see and why. |
| Loop Engineering profile | npm run -s runtime:e2e:loop-engineering-profile | Loop iterations can be observed, guided, attributed, and measured. |
Generated Artifacts
| Artifact | Contract |
|---|---|
| SDK quickstart | aionis_sdk_quickstart_result_v1 |
| HTTP quickstart | aionis_http_quickstart_result_v1 |
| Multi-agent quickstart | aionis_multi_agent_quickstart_result_v1 |
| Memory Firewall quickstart | aionis_memory_firewall_quickstart_result_v1 |
| Memory Firewall A/B demo | aionis_memory_firewall_ab_demo_result_v1 |
| Claude Code MCP demo | aionis_claude_code_mcp_demo_result_v1 |
| Claude Code real transcript | real MCP connection transcript |
| Flight Recorder incident demo | aionis_flight_recorder_incident_demo_result_v1 |
| Loop Engineering profile | aionis_loop_engineering_profile_result_v1 |
What To Look For
When you inspect the generated JSON, check for these surfaces:
| Surface | Why it matters |
|---|---|
guide_trace_id | Proves feedback can be tied to the exact guide exposure. |
agent_context | Shows what the Agent actually receives. |
memory_use_receipt | Shows why memory was used, suppressed, or rehydrated. |
memory_decision_trace | Shows the operator-level decision chain. |
effect_report | Shows whether Aionis can claim a measured product effect. |
flight_recorder_report | Shows after-incident replay without raw prompt dumping. |
What These Examples Do Not Claim
They do not claim that Aionis solves arbitrary external coding tasks by itself. Aionis is not an autonomous runner. It does not replace the Agent host, shell, browser, CI, or verifier.
They show the part Aionis owns:
memory evidence -> governed context -> attributed outcome -> measured effectMinimal Agent Source
The shortest complete TypeScript host example is:
docs/examples/minimal-agent.ts
Use it as the reference shape for your own Agent loop.