Linear · ACME-214Add CSV export to the invoices page
- Condux · Picking up ACME-214. The issue is ready as written.
- Condux · Preview is up at preview-3f1a.acme.dev.
- Condux · Opened PR #1287. Moved to In review.
Hand us an issue the way you’d hand it to an engineer.
Condux writes the code, tests it in a real browser, has it reviewed, and opens one clean pull request. If anything in the issue is unclear, it asks first.
First 5 hours free. No card required. By signing in you agree to the Terms and Privacy Policy.
Linear · ACME-214
GitHub · acme/app · #1287Add CSV export to the invoices page
- 1 commit · +214 −31 · 6 files
- CI passed · preview deployment active
- QA report and review notes in the PR body
“It asked a question I hadn’t considered, the kind a good engineer would catch. The whole thing took me about ten minutes, and I was comfortable merging it and deploying it to production.”
IanCo-founder, Kestrel
Kestrel
KestrelWorks with
- Linear
- GitHub
Slack
Runs your agents on
OpenAI
Anthropic
OpenRouter
Before you ever see a pull request
Every delegated issue passes through six specialists, each built and prompted for one job.
Reads the issue against your codebase, finds what’s ambiguous, and asks you one question at a time until it’s ready to build. Ready issues go straight through.
Codes the change on its own branch in a secure, isolated environment set up for your repository.
Deploys the change to a temporary environment using your own deploy process, so it can be used before it’s reviewed. Gone when the pull request closes. Production is never touched.
Uses the preview in a real browser the way the issue describes, checking for regressions and accessibility problems along the way. What it finds, with screenshots, goes straight to the reviewer and into the pull request.
A separate agent, on a separate model, reviews the code and the QA findings, prompted to find defects and security problems rather than defend the work. It fixes what it finds and refines the change.
One commit, one pull request, the issue moved to review. Your engineers merge, or your merge rules do.
Modeled on how Stripe’s Minions and Cloudflare’s AI code review run coding agents at scale: specialists in sequence, checked by agents that didn’t write the code, and a merge-ready pull request at the end.
It all runs on Condux
Every agent works in its own isolated environment on Condux’s infrastructure. You connect a repository and delegate an issue. Condux does the rest.
- As many agents as you have issues
- Each in its own isolated environment. Delegate ten issues and ten agents start.
- An environment for your repository
- Built when you connect it, kept warm, rebuilt when your setup changes.
- Your models, your subscriptions
- Condux runs on the ChatGPT or Claude plan you already have, or your own keys. Model usage is billed by your provider.
- A record of every run
- Every stage’s transcript, console, and cost, kept for you to read at any time.
Your tools, easily connected
- GitHub
Connect a repository. Condux works on a branch and opens pull requests there.
- Linear
Delegate an issue to Condux. Questions and progress come back on the issue.
- Slack
Talk to Mason, Condux’s product manager, in the channels your team already uses.
- OpenAI
Use your ChatGPT subscription or OpenAI API key for Condux’s agents.
- Anthropic
Use your Claude subscription or Anthropic API key for Condux’s agents.
- OpenRouter
Use your OpenRouter API key for Condux’s agents.
- Coming soon
SentryGive agents production-error context for evidence-backed fixes.
- Coming soon
GitLabConnect a GitLab repository. Condux opens merge requests there.
- Coming soon
Help ScoutTurn customer conversations into product work.
Contained, from first commit to pull request
- Only the repositories you choose
- The GitHub App is installed per repository, with permission to read, push branches, and open pull requests.
- Access lasts one run
- Repository access is minted for each run and expires with it. Your keys and secrets are encrypted at rest and reach a stage only while it runs.
- Every run in its own environment
- Each stage runs in a fresh, isolated environment built for your repository. Workspaces are fully separated from one another.
- Your merge rules decide
- Condux opens the pull request. Your engineers or your branch protection merge it. Request changes and Condux picks them up on the same branch.
Common questions
- What do I need to do to set up a codebase?
- Connect the GitHub repository. Condux’s setup engineer reads the codebase, your CI configuration, and any devcontainer, then opens a pull request that configures the project for Condux. Review it, merge it, and the repository is ready for its first issue.
- If I want Condux to change something on the pull request, how do I tell it?
- The same way you’d tell a colleague: request changes in the GitHub review. Condux reads the review, makes the changes on the same branch, and replies on the pull request.
- What if my Linear issues aren’t well written?
- The product manager reads each issue against your codebase and asks you one question at a time until it’s ready to build. For issues that are still rough, mention Mason in Slack. He works the idea through with you and writes the issue.
- What kind of issues should I give it?
- Anything a good engineer could pick up and finish from the issue alone: features, fixes, refactors, migrations, integrations. Work that needs a product decision mid-flight gets a question. Work that needs an architecture debate is better started in Slack with Mason.
- Does it respect my branch protection, required reviews, and CI?
- Yes. Condux opens a normal pull request. Your CI runs on it, your required reviewers are still required, and your merge rules decide when it lands.
- What happens if the pull request gets merge conflicts?
- Condux owns its pull request until it merges. When the base branch moves and the pull request conflicts, Condux resolves the conflict and pushes the merge, keeping it ready to merge. When a conflict needs a human decision, it says exactly why on the pull request.
- How long does a run take?
- Most issues finish in 40 to 90 minutes of run time. Model usage goes to your own provider at your rates.
- What happens when I run out of hours?
- Your first 5 hours are free, no card required. After that, pick a plan. On a plan, runs keep going past the included hours at the plan’s rate for additional hours, $8 on Starter and $6 on Team, billed at the end of the month. A run already in progress is never stopped.
- How do I cancel?
- From Billing in your workspace settings, any time. Your plan stays active until the end of the current billing period, and you pay only for additional hours used before then. Your runs, pull requests, and history stay in Condux.
- How is this different from Devin, Copilot’s coding agent, or Codex?
- Those give you one agent’s attempt. Condux runs a process: a product manager who checks the issue first, an engineer, a preview, a QA engineer in a real browser, and a separate reviewer on a separate model, all before the pull request opens. You also bring your own model plan instead of paying a markup.
Pricing
Starter
$79per month
- 10 hours of run time a month
- $8 per additional hour
- 1 codebase
- 2 runs at a time
- Full run history and transcripts
Team
$199per month
- 30 hours of run time a month
- $6 per additional hour
- Unlimited codebases
- Unlimited runs at a time
- Full run history and transcripts
Your first 5 hours are free. No card required.
Run time is active time. Waiting on you, on review, or on CI doesn’t count.
Model usage is billed by your provider at your rates.
By signing in you agree to the Terms and Privacy Policy.
Ready for your first Linear issue within the hour.
Connect GitHub and Linear and choose a repository. Condux opens a pull request that sets the codebase up. Merge it, and delegate your first issue.
Sign in with GitHubBy signing in you agree to the Terms and Privacy Policy.