Skip to main content
Sekreton
Sign in Start free trial

You approve plans and results, agents do everything else

Sekreton runs your coding agents in parallel, each in its own cloud sandbox. They read the codebase, draft the plan, build, test, and clear their own merge conflicts and red CI. Nothing runs before you approve the plan; nothing merges before you approve the result.

Start your 7-day free trial See how it works

7-day free trial · Cancel anytime · Bring your own model key

Sign in with Google or GitHub

A team board in Sekreton, with tasks moving from To Do through In Progress and In Review to Done

Works with GitHubGitLab AnthropicOpenAIMoonshotOpenRouterOpenCode ZenOpenCode Go

Two decisions per task. Both yours.

Everything between the brief and the merge happens without you. You only see a task when it needs a decision.

  1. Brief the work

    A sentence or a page, written yourself or drafted by the assistant from a chat.

  2. Approve the plan

    The agent reads the codebase and proposes an approach. Approve it, or send it back with notes.

  3. Agents do the work

    They build, run your lint and test gates, and fix whatever fails, including merge conflicts.

  4. Approve the result

    Read the diff inline, comment on lines, request another pass, or merge. One squashed commit per task.

The plan reaches you first

Before a line is written, you see what the agent intends to change and why. A bad idea costs you thirty seconds of reading, not an afternoon of review.

  • Approve, or reply with notes and the agent revises
  • Every plan revision kept side by side
  • Research and writing tasks deliver documents, not diffs
A task in Sekreton showing the agent’s four-step plan to fix a double charge on slow networks, waiting for approval
Sekreton’s diff review: the checkout fix with its verification gates passed, a reviewer’s inline comment and a Merge button

Results arrive green

Merge conflicts, failing lint, red tests, a base branch that moved overnight: that’s the agent’s problem, not yours. By the time a diff reaches you, it passes your gates on top of the latest base.

  • Your own lint and test commands are the gate
  • Conflicts with sibling tasks rebased and resolved
  • Comment on a line and the agent takes another pass

The leverage of a team, without the headcount

A sandbox for every task

Every agent and every check runs isolated in its own secure container and branch, so nothing touches your machine or another task.

Every project at once

Run as many tasks as you have ideas. Nothing queues behind you.

Approve from your phone

It all runs in the cloud. Sign off a plan from the queue, a diff on the train.

Your model, your bill

Bring a key from Anthropic, OpenAI, Moonshot, OpenRouter, OpenCode Zen or OpenCode Go. Use cheap models where they’re good enough.

Nothing hidden

A live stream, full transcripts and usage for every task.

Built for teams

Seats and roles, and an MCP connector so your own tools can file and read tasks.

Work that starts itself

Put an agent on a schedule, or on your support inbox. It works while you sleep and brings you a result to approve.

See every recipe

Simple, flat pricing

No per-seat fees. Model usage is billed by your own provider.

Individual

For one developer with more projects than hours.

$20 USD / month

  • 2,000 compute-minutes a month
  • One member
  • Unlimited projects and tasks
  • Extra minutes $10 per 1,000
Start 7-day free trial
Unlimited members

Team

One flat price, however many people review.

$49 USD / month

  • 10,000 compute-minutes a month
  • Unlimited members, seats and roles
  • Unlimited projects and tasks
  • Extra minutes $10 per 1,000
Start 7-day free trial

A compute-minute is a minute of sandbox time, whether an agent is working or your lint, tests and CI are running. Plus GST or VAT where it applies. Plans and diffs stay readable without a subscription.

Questions

Something not covered? Write to hello@sekreton.com and a founder answers.

Does anything ship without me?

No. A task stops twice and waits: once for you to approve the plan, once for you to approve the result. Nothing reaches your base branch that you haven’t read.

Where does my code go?

Into a container of its own for the length of one task, checked out from your forge and destroyed with it. Prompts and code excerpts go to the model provider you configured, under its terms. Read the privacy policy.

What happens when two tasks collide?

Each task has its own sandbox and branch. When a merge conflict or a broken gate does turn up, the agent resolves it and comes back with a diff.

Which agents and models can I use?

Bring a key from Anthropic, OpenAI, Moonshot, OpenRouter, OpenCode Zen or OpenCode Go. Your provider bills the model usage; Sekreton bills only compute-minutes.

Who gets the most out of it?

Anyone whose job is reading plans and reviewing diffs: a solo developer with more projects than hours, or a team that wants one person’s judgement applied across all of them.

Brief the work, approve the plan, judge what ships. Let agents do the donkey work.

Start your 7-day free trial