syneva Get started

Code fast.Ship with confidence.

Your agent moves fast. Understanding its work shouldn't slow you down. Make sense of the plan. Review the changes. Decide what ships.

Start a local review See how it works
YOUR AGENT'S DIFF. YOUR ATTENTION.
The review, in two dimensions The full diff passes from your agent into a local review. Its layers open vertically: critical, important, and tested. You make your decisions during review, then ship code you've examined with confidence. A dotted return connection carries the next revision back to the agent. An illustrative human review, not automatic approval or a product screenshot. Your full diffEvery change from your agent TestedCheck evidence ImportantFollow the change YOUR FOCUS CriticalRead closely Ship with confidenceCode you've reviewed SAME TAB. NEXT REVISION.
Your full diff. Room to understand it. Confidence in what ships.
Local by defaultYour agent, not oursOpen source · MITNo model runs inside Syneva

Writing code got faster.
Knowing what to trust didn't.

A long plan. A bigger diff. Another round of scrolling, guessing, and asking your agent what changed. Syneva is the layer between what your agent produces and what you ship: a place to do that work, so your attention goes to the decisions only you can make.

The diff got bigger.
Trust didn't come with it.

Nobody reviews a thousand lines with the attention a hundred deserve. Syneva is the layer that closes that gap: it makes every change readable, questionable, and decidable — by you, not by a model. The numbers below are product facts, not growth metrics.

0changes auto-approved. Every verdict is a recorded human decision.
1tab carries the whole loop — review, questions, and the next revision.
4review modes: working tree, staged changes, a branch or PR, a single file.
100%local. No model inside, no telemetry, no account. Your machine, your code.

A clear plan.
A better starting point.

Plan desk · prototype

Don't lose the important decision in a wall of markdown. The plan desk we're exploring turns an agent's proposal into a map you can question.

  • See the shape of the work.Read the plan as a map, a document, or source.
  • Find the assumptions worth checking.See which claims carry risk, and what would prove them wrong.
  • Give your agent a clearer brief.Comment on the exact line. Settle the decision before the code.

This is the direction, not a feature available in the CLI today.

A plan becomes a map of decisionsA plan document routes into three claims. One needs proof, one is a question for the agent, and one is understood. Illustrative prototype concept. plan.md AssumptionNeeds proof A questionAsk the agent A decisionUnderstoodLESS READING BLIND. MORE DECIDING.
Plan → claims → decisions

A big diff.
A smaller mental load.

Available now

The agent has finished. Now it's your turn. Review its work in a local browser desk, with a reading order and a direct line to the agent that wrote it.

  • Follow the change, not the file list.Your agent can group files into a walkthrough. Every file stays visible.
  • Ask right where it matters.Questions and change requests stay anchored to the code.
  • Keep the good. Send the rest back.Accept or reject changes. Send your verdict. Review the revision in the same tab.
Try it on your working tree
A reading order through agent-written codeThree files come in a guided reading order. A question is anchored to a changed line, the agent answers it, and the human records a verdict. Illustrative review, not a screenshot. ? auth/session.ts api/middleware.ts+ verifySession(request)? YOU · Why here?AGENT · Before the handler. routes/account.tsAccepted by youONE TAB. THE WHOLE REVIEW LOOP.
Read → question → decide → revise

Confidence is built
round after round.

01Open the desk

Run syneva next to your repository. The diff opens on localhost, in your browser — nothing leaves your machine.

02Read with a guide

Your agent attaches a reading order and groups the files into sections. Every file stays visible, listed or not.

03Ask on the line

Questions anchor to the exact change. Your agent answers in the same thread, without you leaving the desk.

04Send the verdict

Accept or reject each change. Your agent receives a structured result and re-diffs its edits into the same tab.

Anything your agent rewrites resets to pending. Anything you settled stays settled.

The verdict,
as a contract.

Send to Agent hands back a structured ReviewResult — printed to stdout and written to disk, readable by any harness. Not a screenshot, not a vibe: your agent knows exactly what you accepted, what you rejected, and what you asked.

It acts on the round, calls syneva reload, and the same tab shows its next revision — your settled decisions carried over, its edits back to pending.

Read the full contract
SYNEVA SPEC · REVIEWRESULT
{
  "mode": "repo",
  "accepted": [{
    "path": "auth/session.ts",
    "title": "Verify session before handler"
  }],
  "rejected": [{
    "path": "api/middleware.ts",
    "title": "Inline session check"
  }],
  "requestedChanges": [{
    "path": "api/middleware.ts",
    "body": "What happens when the session expires?"
  }],
  "approvedFiles": ["routes/account.ts"],
  "openQuestions": []
}
Illustrative round. The full shape is versioned with the CLI: syneva spec

Your code stays
on your machine.

Review doesn't need another cloud service. Syneva opens next to your repository, on localhost. It doesn't call a model or replace your agent. It gives you a better place to judge the work.

Recorded performance evidence
Historical runs published September 17, 2026, not a guarantee for your repository.

localhostYour repoYour verdict

Keep the speed.
Keep the understanding.

Open a desk over your working-tree diff. Bring the agent you already use.

Read the setup guide

Then, inside your repository

syneva

A few things
worth knowing.

Does Syneva review code for me?

No. You review the code. Your own agent can organize the reading order and answer questions. Syneva records your decisions and hands your feedback back to that agent.

Does Syneva ever auto-approve anything?

No. Every change starts pending, and only your click records a decision. A decision is a verdict stored on that exact change — not git staging, not a blanket approval of a file.

What if the agent rewrites code I already accepted?

Syneva hashes the content of each change. On reload, a rewritten change resets to pending; untouched work keeps your verdict. Approvals and comment anchors follow the same rule, so stale trust can't survive silently.

Which coding agents can I use?

The CLI uses a documented JSON and localhost HTTP contract. It includes a pi integration, and other agents can use the standalone CLI. See the setup guide for the commands your agent needs.

Can I use the plan desk today?

The dedicated plan desk is a prototype. The current CLI can open a file for review, including a plan, but the claim map and risk-ranked queue shown here are not shipped yet.

Where does my review live?

On your machine, under ~/.syneva. The desk binds to localhost by default and saves decisions between rounds. There is no hosted account or telemetry.