AegisOpsLive enterprise agent systems lab
live public data GitHub Models free read-only Postgres-ready Source
ready
SRE / incident response3 model agents

Production Incident Investigator

Investigate the live GitHub platform status, reconcile component health and active incidents, then produce an evidence-grounded operational brief.

GitHub Status API + NIST policy corpus
Service scope: github-platformTime window: current
Live execution canvas

Agent decides. Rules match.

0 agent tools 1/6 shown · 1/6 ready
Live execution story

Step through the same decision in both systems

Stage 1 of 6 · 1 live
Agentic executionObserves, decides, acts, and can adapt
Evidence can change the next step
Stage 1 · HUMAN + LANGGRAPHBound the goalqueued
Outcome targeted in this stageAccept a bounded case and enforce its tool, budget, and side-effect limits.
Inputs available
Service scopeScenario inputgithub-platform
Time windowScenario inputcurrent
Decision or actionSafety and autonomy boundary

Typed scope, tool allowlist, spend ceiling, and approval requirement are checked before execution.

Why this step matters

The goal and safety boundary stay human-controlled; autonomy starts only inside that approved boundary.

Stage result
configured toolsgithub_status, github_incidents, enterprise_policy_search
Planned stage · run to observe the actual decision
Deterministic executionValidates, derives, decides, and audits exactly
Strong inside its encoded world model
Stage 1 · ZOD + REGISTRYValidate decision contractqueued
Outcome targeted in this stageValidate the versioned decision contract and safe exception path.
Inputs available
Service scopeScenario inputgithub-platform
Time windowScenario inputcurrent
Decision logicVersioned boundary

The input schema, source allowlist, rule version, and approved outcomes are validated before any source is called.

Strength and boundary

This is a production strength: every permitted input and outcome is explicit, testable, and reproducible. Its boundary is the model encoded here.

Stage result
approved sourcesgithub_status, github_incidents, enterprise_policy_search
fallbackmanual exception review
Planned stage · run to observe the actual decision
Live topologyEvery lit node is backed by the same received trace shown above.
Agentic laneadapts plan and tools
Deterministic lanederives facts and evaluates decisions
React Flow
Outcome comparison

Same live inputs. Different capability boundary.

AGENTIC OUTPUTGrounded adaptive conclusion

Run both lanes to stream the agent’s cited conclusion here while its trace is still visible above.

Parallel specialists reconcile independent evidence and adapt the operational brief to what they observe.
DETERMINISTIC OUTPUTConfigured outcome match
No result yet

Run both lanes to evaluate the versioned rules against the same live facts.

Thresholds route known severity states; they cannot investigate conflicting or novel evidence combinations.
Why agentic for this caseParallel specialists reconcile independent evidence and adapt the operational brief to what they observe.

Use rules for the known, repeatable boundary. Use the agent only where evidence can be ambiguous, the next useful action depends on what was just observed, or multiple findings must be reconciled.

Public demo charge$0.0000GitHub Models free tier
Direct API equivalent$0.00000Measured input + output tokens
Deterministic model cost$0.0000json-rules-engine, no LLM