No team should re-solve
the same fire twice.
helena indexes every past fix from your team’s incident channels, dashboards and error trackers. When a new alert fires, it surfaces the resolution your team already worked out. Powered by BTL Runtime with per-role model routing.
One slash command.
Cited answers in seconds.
Type /askoncall in Slack or Discord and helena finds the past fixes, ranks them, and drops a runbook style card back into the thread with every incident id it cited.
- Reranked candidates, not the first FTS hit
- Every claim tied to a real INC id
- No hallucinated commands, only ones your team ran before
See the reasoning,
not just the answer.
Every Copilot turn streams a live trace: which model classified, which tool got called, which incidents matched, and the citation validation step before the answer ships. Nothing is hidden.
- SSE events for classified, tool_call, tool_result, delta, final
- Clickable INC and RB pills open the source in a new tab
- Attach a screenshot, two vision models must agree
Every source lands in the
same incident memory.
Grafana alerts, Sentry errors, GitHub deploys, and generic webhooks feed the same table. Severity colored, source tagged, dedup hashed. Click any row to see the full body and every Copilot answer that has ever cited it.
- Dedup on source + external_id so retries do not spam the feed
- Postgres tsvector search under 50 ms as the corpus grows
- Deployment rows show commit title, author, diff stats
Every token, every role,
on one page.
Classifier, reasoner, rerank, synth, and vision each track input and output tokens per turn. Costs sum against BTL Runtime prices so you always know what a Copilot answer or a runbook draft actually cost. No surprise bills at month end.
- Per role breakdown so cheap models stay on cheap jobs
- Rolling 30 day view with per turn detail on hover
- Fallback model calls flagged separately when routing kicks in
Three steps. Under two minutes.
Ask once. helena reasons across every source.
Pick a question. Watch which sources helena decides to query and how the answer comes together.
