Skip to content

use-cases.md

Latest commit

 

History

History
115 lines (79 loc) · 3.32 KB

use-cases.md

File metadata and controls

115 lines (79 loc) · 3.32 KB
Raw

Use cases

Practical ways teams put the toolkit to work. Every workflow below is read-only — nothing here runs a remediation or mutates infrastructure.

1. Fast incident triage from the terminal

Pipe the most relevant output straight into the analyzer to get a ranked hypothesis and read-only commands to confirm it:

kubectl logs my-pod --previous | devops-ai analyze - --tech kubernetes

The result tells you the most likely cause, the diagnostic commands to verify it, and a suggested fix — without you having to remember the right kubectl incantation.

2. On-call runbook lookups

Instead of digging through wikis, ask the catalog directly:

devops-ai explain CrashLoopBackOff
devops-ai explain "exit code 137"
devops-ai explain "Error acquiring the state lock"

Pair it with the hosted AI incident assistant for live escalations.

3. CI gate on known failure patterns

Use the exit codes to fail fast when a build trips a catalogued error:

terraform plan 2>&1 | devops-ai analyze - --tech terraform
if [ $? -eq 0 ]; then
  echo "::warning::Known Terraform failure pattern detected"
fi

analyze exits 0 on a match, 1 on no match, 2 on input errors. See the CLI guide.

4. Pre-deploy manifest validation

Catch problems before they reach the cluster, read-only:

devops-ai validate deploy.yaml
devops-ai validate main.tf

Wire it into a pre-commit hook or CI step.

5. Programmatic enrichment in your own tooling

Embed the engine in dashboards, bots, or internal tools via the SDK:

from devops_ai_toolkit import AnalysisEngine

engine = AnalysisEngine()

def triage(log_text: str) -> dict:
    result = engine.analyze_text(log_text, enrich=True)
    return {
        "summary": result.summary,
        "confidence": result.confidence_percent,
        "top_fix": result.suggested_fixes[0].title if result.suggested_fixes else None,
    }

6. A shared HTTP service for your platform team

Run the REST API once and let every team query it:

devops-ai serve --host 0.0.0.0 --port 8000
curl -s platform-svc:8000/analyze/log \
  -H 'content-type: application/json' \
  -d '{"content": "ImagePullBackOff"}' | jq .summary

Same engine, same results as the CLI and SDK. See the REST API guide.

7. ChatOps integration

Forward a Slack message or alert payload to the engine and post back the summary plus the top diagnostic command. Because output is structured JSON, it's trivial to format for chat.

8. Private / air-gapped analysis

The deterministic engine needs no network. For environments that still want an LLM narrative without sending data to a vendor, point it at a local Ollama instance:

export DEVOPS_AI_PROVIDER=ollama
devops-ai analyze incident.log --enrich

9. Building your own knowledge base

Capture your organization's recurring incidents as signatures and inject a custom knowledge base — turning tribal knowledge into a deterministic, queryable catalog. See the Plugin guide.

See also