Provenant by Roveldi
Every incident answered with evidence, not guesswork.
Provenant is the operational intelligence layer for engineering teams. It investigates production incidents across your logs, metrics, deployments and code, delivers an evidence-backed root-cause analysis, and learns from every outcome your engineers confirm.
Incident INC-2041 · RCA
Confidence: High
Root cause: connection-pool exhaustion after deploy a1f3c9e
payment-service · pool size reduced 50 → 10 in config change
- Loki
- Prometheus
- GitHub
- Deploy
- Pool wait time p95 rose 40× at 10:42Fact
- Errors began 3 min after deployObservation
- RabbitMQ backlog is secondaryHypothesis
- Roll back config (needs approval)Recommendation
The problem
Your operational knowledge is everywhere — except where you need it.
01
Tribal knowledge
A few senior engineers carry most of the incident context.
02
Scattered signals
Logs, metrics, deploys, runbooks and chat are stitched together manually under pressure.
03
Repeated investigations
Similar incidents get investigated from scratch — and lessons leave with people.
- 57%
said their most recent major outage cost over US$100,000.
Source: Uptime Institute, Annual Outage Analysis 2026
- 4 in 5
said their most recent serious outage could have been prevented.
Source: Uptime Institute, Annual Outage Analysis 2024
What Provenant does
From alert to trusted root cause.
Investigate
Bounded, read-only tools gather evidence from your observability stack, deployments and repository.
Correlate
Runtime signals are linked to the deployed commit, code and config changes, and service dependencies.
Explain
Ranked hypotheses become an RCA with cited evidence, confidence, blast radius and recovery status.
Remember
Engineer-confirmed resolutions become trusted institutional memory for the next incident.
Why Provenant
Built on trust, not just a model.
Evidence before assertion
Every conclusion links to its source evidence.
Facts are not hypotheses
Each statement is labelled fact, observation, hypothesis or recommendation.
Deterministic numbers
Impact counts are calculated, never generated by the model.
Human in control
No remediation executes without explicit approval.
Model-agnostic
Run locally or with your preferred LLM provider, with full provenance.
Validated learning
Only human-confirmed outcomes become institutional memory.
Product
See the reasoning, not just the answer.
- Deploy a1f3c9e to payment-serviceDeploy
- Pool wait time p95 spikesPrometheus
- “Timeout acquiring connection” × 412Loki
- Settlement queue backlog growsRabbitMQ
- Config diff: pool.max 50 → 10GitHub
- Connection-pool size reduced in config changeSupported
- Downstream bank API latencyWeak evidence
- RabbitMQ consumer crashWeak evidence
Every AI response records the provider, model and configuration that produced it. Illustrative data.
Integrations
Works over the stack you already run.
Not another observability tool — the reasoning and memory layer on top of yours.
Available
On the roadmap
Who Provenant is for
SRE & on-call
Narrow a production problem faster with evidence already assembled.
Engineering leaders
Consistent, auditable RCAs and visible investigation quality.
Platform teams
Keep operational knowledge when people move on.
Built for payments, fintech, CPaaS and transaction-heavy SaaS teams with 50–500 engineers.
Prove it on your own incidents.
A 4–6 week Proof of Value replays your historical incidents, observes live investigations in advisory mode, and ends with a quantified business review against metrics we agree upfront.