Global Technology Partners
← All accelerators
LangGraphAnthropic ClaudePython

LangGraph AIOps Accelerator

Incident response with a human approval gate

Incident response is a good fit for agents only if they stay inside guardrails. This build gathers context and proposes a fix automatically, then stops and waits for an on-call engineer before anything runs.

Benefits

Approval gate

The graph pauses with a LangGraph interrupt and resumes only when an engineer approves or rejects.

Constrained actions

Only actions on an allowlist can ever run, and execution defaults to dry-run.

Knows when not to act

Low-confidence diagnoses and informational alerts are reported for review and never trigger action.

Pluggable

Swap the fixture tools for your monitoring and orchestration stack, and swap the diagnoser between Claude and a rule-based one.

How it works

  1. 1.Alert received
  2. 2.Gather logs, metrics and recent deploys
  3. 3.Diagnose root cause with a confidence score
  4. 4.Propose an allowlisted action
  5. 5.Pause for on-call approval (interrupt)
  6. 6.Execute in dry-run, then report

How to use it

Install

pip install -e .[dev]

Run the tests

pytest

Offline run, pauses for approval

python -m aiops_agent.cli --eval

Demo: approve everything (still dry-run)

python -m aiops_agent.cli --eval --auto-approve

Diagnose with real Claude

python -m aiops_agent.cli --eval --live

Offline demo: four synthetic incidents

Captured from a real run with the offline rule-based diagnoser and automatic approval. Actions are dry-run only. This validates the graph, approval gate and allowlist, not diagnostic quality.

$ python -m aiops_agent.cli --eval --auto-approve
[CRITICAL] checkout-api: p95 latency above 2s
Root cause: db_pool_exhaustion (confidence 85%)
Executed after approval by demo: [dry-run] would run scale_db_pool on checkout-api

[CRITICAL] search-service: 5xx error rate above 10%
Root cause: bad_deploy (confidence 80%)
Executed after approval by demo: [dry-run] would run rollback_deploy on search-service

[WARNING] report-worker: CPU above 95% for 10 minutes
Root cause: capacity_saturation (confidence 75%)
Executed after approval by demo: [dry-run] would run scale_out_service on report-worker

[INFO] auth-service: disk usage above 90%
Root cause: unknown (confidence 30%)
No automated action taken; left for on-call review.

root_cause_accuracy=100% action_accuracy=100% (n=4, synthetic)

Read the results honestly

  • •The 100% score is a keyword-rule diagnoser on 4 synthetic scenarios. It does not measure diagnostic quality or MTTR.
  • •Production use needs read-only credentials first, narrow action scopes and an incident-review process.
  • •Tools are fixture-backed until replaced with your monitoring and orchestration integrations.

Reference build demonstrated on synthetic data, not a client deployment. Names of third-party products describe what the build runs on and do not imply endorsement or partnership.

Want this adapted to your environment?

We tailor it to your data, systems and approval process, and measure results on your own cases.

Talk to Us