Commercial observability for AI-assisted engineering

Know what AI-assisted work actually ships.

Clavinci joins AI usage, git output, Jira delivery, and PR collaboration. It measures cost and outcomes without reading prompts, responses, source code, or CI logs.

The joined evidence trail

live model / 04 joins
  1. 01AI usage
  2. 02git output
  3. 03Jira delivery
  4. 04PR collaboration
joinedaudited

Four structured sources join into one attributed and audited evidence seam.

Joined evidence

Four streams. One evidence trail.

Tokens alone are spend. Commits alone are activity. Issues alone are delivery. Clavinci connects them so teams can ask about outcomes without turning engineers into a leaderboard.

01

AI input

Captured
Token counts, models, sessions, cache, and tool metadata.
Never
Prompt or response content.
02

Git output

Captured
Commit metadata, file paths, LOC deltas, CI timing, and deployment markers.
Never
Source code or CI logs.
03

Delivery

Captured
Jira issue metadata, sprints, transitions, and story points.
Never
Descriptions or comments.
04

Collaboration

Captured
Pull request and review events, turnaround, and iterations.
Never
Review comment text.

Outcome proof

Spend is not the answer. Shipped work is.

Clavinci attributes AI sessions to commits and issues, keeps the join replayable, and reports the measures that can change an engineering decision.

  1. 01

    Cost per merged LOC that survived 30 days

  2. 02

    Dark sessions with spend and no outcome

  3. 03

    AI lead-time delta

  4. 04

    Sprint spillover by attribution

  5. 05

    PR review health

  6. 06

    Tokens per completed issue

Every measure is derived from structured metadata. The raw work stays out of the system.

Trust contract

The trust model is the product.

Engineers keep the agent installed only when the collection boundary and the access trail hold up under review.

01

Engineer

Your own history, available locally. Your audit log. Your attribution disputes.

02

Direct manager

Individual views for direct reports. Every individual read is logged.

03

Skip-level

Team aggregates by default. Individual access requires a typed reason and leaves a visible audit row.

04

Admin

Coverage, retention, model prices, Jira configuration, and operational health.

Anti-metrics

Useful evidence needs a refusal list.

Clavinci blocks metrics that reward typing volume, invite public ranking, or automate management decisions.

Not measured

  • time on keyboard
  • public leaderboards
  • LOC ranking
  • comments per PR ranking
  • estimate accuracy by engineer
  • automated people decisions

The blocklist is enforced at the query layer, not hidden in the interface.

Operation

A hosted product with a strict data boundary.

Clavinci is operated by the company as a hosted service. Local collectors capture structured metadata and send only the allowed event contract. Customers do not deploy or maintain the application stack.

Company-operated service. Managed updates. Passive collectors. Bounded collection. Soft-fail integrations.

01 / collectors

Eight AI tools, one collector contract

  • Claude Code
  • Codex CLI
  • Cursor
  • Gemini CLI
  • Kiro
  • OpenCode
  • OpenClaw
  • Every Code

02 / outcomes

Delivery and operations sources

  • git
  • GitHub Actions
  • GitLab CI
  • Jira
  • PagerDuty
  • Opsgenie
  • incident.io

Product evaluation

Measure the work. Keep the trust.

Review the trust model, commercial offer, and comparison records, then decide whether Clavinci fits your organization.