Who this is for
AgentOps Flight Recorder is for SRE, platform, security, and operations teams that let AI agents touch tickets, tools, APIs, dashboards, and release workflows.
The problem
AI agents can move fast, but after a risky run a chat transcript is not enough. Teams need to answer: what did the agent do, which tools did it call, where did risk appear, who approved the next step, how much did it cost, and what can safely happen next?
That is the operational gap this project solves.
What it does
AgentOps Flight Recorder turns AI-agent work into Splunk-ready evidence:
- structured AgentOps event records
- Splunk HEC-shaped JSONL exports
- SPL search drafts for timeline, risk, approval, retry, and cost review
- a working dashboard that filters risky actions and approval gates
- an evidence-bound AI investigation draft that cites event IDs instead of inventing facts
- a handoff path for humans to approve, reject, or resume the work
How AI is used
AI is not used as an unaccountable black box. In the demo, the AI investigator is constrained by the event trail. It can summarize the incident only from cited event IDs and SPL-style result rows. If evidence is missing, the workflow keeps a human approval point instead of pretending the answer is complete.
Working demo
The updated demo video shows the project functioning in the browser. It loads the event trail, filters risky actions, shows human approval gates, runs a Splunk-style search, and generates the evidence-bound AI investigation draft.
Demo video: https://youtu.be/Hg1QRVr76Bs Live demo: https://daideguchi.github.io/agentops-flight-recorder/ GitHub: https://github.com/daideguchi/agentops-flight-recorder
How it was built
The project includes a Python event pipeline, shared AgentOps JSON schema, Splunk HEC-shaped payloads, SPL query drafts, a static working dashboard, repeatable verification scripts, and a narrated working-demo video.
Verification command:
cd splunk-agentic-ops
bash scripts/run_splunk_local_checks.sh
Observed proof:
splunk_local_checks_ok
hec_events=26
video_seconds=53.8
architecture=root
claim_boundary=verified_local_splunk_ready_no_live_ingestion_claim
Claim boundary
This submission proves a verified local prototype and Splunk-ready event package. It does not claim live Splunk Cloud ingestion yet.
Value
If AI agents become part of operations, their work needs observability. AgentOps Flight Recorder makes agent activity searchable, explainable, and reviewable before teams trust it with more responsibility.
Log in or sign up for Devpost to join the conversation.