Skip to main content

Review and evidence for Claude Code and Codex

Review AI coding work without rereading the session.

Jentrix turns each agent task into a structured record of the request, plan, decisions, outputs, failures, and unresolved gaps. Review the evidence, then accept or return the work.

Request free accessView a real task record

Agent sessions become reviewable task evidence.

Works with Claude Code by Anthropic and Codex by OpenAI.
task-centred evidencehuman acceptance recordedfree for individual developers

Agent work is easy to produce and difficult to review.

An hour of agent work leaves three places to look, and none of them answers the reviewer's question.

The transcriptContains everything — thousands of lines nobody reads twice.
The pull requestContains only the final change, not how the work got there.
The trackerRecords the status, not the reasoning or the gaps behind it.

The same hour of agent work, reviewable.

Without JentrixWith Jentrix
Reread a long transcriptReview a structured task record
Infer why decisions were madeSee decisions and rationale
Trust “task completed”Inspect failures and open gaps
Reconstruct cost and usageSee task-level usage
Approve informallyRecord acceptance or return

from session to board · the record replaces the reread

How it works

01

Connect the task

One prompt in Claude Code or Codex installs the CLI and aligns the session to a card on your board.

02

Capture the evidence

The request, plan, decisions, outputs, failures and unresolved gaps land on the card as typed artifacts. Before Claude Code compacts its context a transcript snapshot lands first, and every run closes with a RUN_SUMMARY.

03

Review the record

Walk the eight categories in order — input to record — or have a reviewer agent walk them first over MCP. The full transcript stays one click away.

04

Accept or return

A human records the verdict, and a return needs a reason. A reviewer agent's verdict lands as one comment on the task — it never moves the card.

The record a session leaves behind.

This is a real card — JEN-224 — as Jentrix keeps it: the request and the plan as typed artifacts, each attributed to the person and the session that produced it, and a review block whose verdict records exactly what you ticked.

input → plan → decision → output → issue → gap → learning → record

A Jentrix task panel for JEN-224: input artifacts (a PRD and the opening prompt) and a plan, each attributed to a person and an agent session; below them a review section with per-artifact read checkboxes, an Accept button and a Return button, and the note that a return needs a reason

Caught before it was accepted.

The agent reported the task complete. The record said otherwise — so the reviewer returned it with reasons instead of accepting on trust.

An example task record — not a customer's

requestAdd retry handling to the payment synchronisation service.agent says“The implementation is complete.”
  • IssueOne integration test remains failing.
  • GapThe retry policy assumes idempotent downstream behaviour.
  • GapA database migration was generated but not executed.
  • DecisionThe error-handling strategy changed after context compaction.
  • RecordThe session's minutes and cost, on the card.

ReturnedResolve the failed test and validate the idempotency assumption.

GitHub shows the change. Jentrix shows how the work reached it.

A connective review layer between the tools you already run. It replaces none of them.

GitHubCode, diffs, checks and pull-request review.
Linear / JiraOwnership, status and workflow.
The transcriptEverything, chronologically — painful to review.
JentrixTask-linked evidence, unresolved gaps and the acceptance decision.

For developers

Finish a coding-agent session with a review-ready record instead of an unstructured transcript.

For engineering leads

Understand what changed, why it changed, and what remains unresolved — before approving the work.

Platform leadership

See how agents are being used, what they cost, and where review risks accumulate.

Your agent work remains your data.

  • Jentrix runs no models and never calls a model provider's API — a repository check proves it on every change. Your agent runs where it always ran.
  • Your keys stay yours. The model bill stays with your provider — Jentrix never charges for tokens.
  • No training on customer content.
  • Recognised secrets are redacted at ingestion — recognised, not infallible, and the record stays private to your workspace.

Free for one person. Access by request, for now.

Ask once; approval lands by email, and from there sign-in is Google or an email link. Every feature, free for one developer — bring your own agent and your own keys, and Jentrix never charges for tokens.

Request free access

no card · no token charges

The questions that decide it

Do I have to change how I work with my agent?

No. One prompt connects the session, then you keep working in Claude Code or Codex as usual — the record accumulates on the card while you do.

Does Jentrix run or proxy the model?

No. Jentrix runs no models and never calls a model provider's API. Your agent runs on your own keys, your provider bills you, and Jentrix keeps the evidence.

What happens to secrets in a session?

Recognised secrets are redacted at ingestion, before storage. Redaction is pattern-based — recognised, not infallible — and the record stays private to your workspace.

Can I start today?

Access is currently by request. Ask once and approval lands by email; from there sign-in is Google or an email link, and the free plan covers one developer with every feature.

The review gate for agent-produced work.

Bring your own agent and your own keys. Jentrix keeps the record — free for one person.

Request free access