Developer: "Tested against the provider's sandbox."
For teams
Your whole team, on the same page as the code.
Technical or not, everyone asks about the product in plain language, sends requests checked against the code, and sees what was delivered, criterion by criterion.
No card needed. Your 14 days start when your first repository is ready.
- Already built? No.
- Contradicts a doc? No.
- Touches: Refunds, Invoices.
Why it's bigger than it looks: Refunds touch invoices, tax records and the payment provider.
Your developer's size: —
- A customer can be refunded part of an invoice.
- The invoice shows the refunded amount.
- The payment provider receives the partial amount.
Developer: "Covered by their test mode."
How to check it yourself: Refund 5 on a test invoice, then open the invoice.
Works with
Claude
ChatGPT
Claude Code
Cursor
GitHub
Notion
Google Docs
Codex
Lovable
In your words
Everyone decides. Not everyone reads the code.
- "I don't know if this should take two days or two weeks."
- "I can't tell if what shipped is what we asked for."
- "Our docs say one thing and the code does another."
- "Requests reach us vague, or citing things that don't exist."
Your first answer
One answer, based on the code, for everyone.
Ask why something takes time or what a change implies. The answer cites the code and carries your developer's note, so the whole team works from the same facts.
Why would changing Refunds be hard?
Refunds touch invoices, tax records and the payment provider. A change has to keep all three consistent.
Features
What your team gets.
Ask, also from Claude and ChatGPT.
Answers cite the code and carry a confidence level. Questions about difficulty say "Ask your developer", so they can add a note.
Requests with an Impact summary.
Checked against the code and docs, with a relative size (S, M or L) and the reason. Never hours. Your developer accepts or edits it.
Delivery Check.
Each accepted criterion as Met, Not met or No evidence, with your developer's answer next to it.
Every change is checked against what you decided.
Every pull request is checked against your docs and rules, with one comment only when it matters.
Know when a change removes or duplicates something that worked.
If a change removes or duplicates something that already worked, you're told.
Roles and one fixed price.
Owner plus up to 10 Developers or Viewers, with no per-seat price and no metric by person.
How it works
Four steps to your first result.
- 1
Sign up.
With GitHub or your email, and two one-click questions about who changes your code.
- 2
Connect GitHub.
You or your developer. Add Notion or Google Docs if you want.
- 3
Get your first results.
Your capabilities, the Retro Check and the first Drift Audit, off your quota.
Invite your team
- 4
Every change, checked.
A plain note for every change, and a warning when something that worked disappears.
Integrations
Use Docdrift in Claude or ChatGPT
Docdrift lives in the chat you already use, in your coding agent and in your AI builder.
01In your chatClaude and ChatGPT
Claude
ChatGPT
Ask Claude or ChatGPT about your product. Docdrift answers from your code.
What you can do from your chat
- Ask about your product. Answers cite the code and say how confident Docdrift is.
- See what your product does, what changed and the open findings.
- Draft a request, see its review and send it to your developer, or confirm it yourself.
- Follow each request through to delivery.
Writes always ask you to confirm in the chat. The assistant inherits your role.
Add Docdrift as a custom connector: in Claude, Customize › Connectors › Add custom connector; in ChatGPT, developer mode › create the app.
Which Claude and ChatGPT plans and devices work
Claude: web, desktop and mobile. Works on Free (one custom connector), Pro, Max, Team and Enterprise. On mobile it uses what you connected on web or desktop. Claude Code also loads the connector with a subscription login.
ChatGPT: on the web, on Plus, Pro, Business, Enterprise and Edu (developer mode). Not on ChatGPT Free or Go, and not in the mobile apps. On some plans ChatGPT can only read: if drafting a request doesn't work there, you send it from Docdrift.
Did my agent remove anything this week?
Yes, one thing. The password reset page was removed on Tuesday, in "Clean up auth screens". Nobody asked for that.
Can my app already take card payments?
Yes. Checkout accepts card payments. It doesn't support bank transfers.
02In your coding agentClaude Code, Cursor and Codex
Claude Code
Cursor
Codex
Your agent checks Docdrift before it deletes anything.
An MCP server (does this already exist? what depends on this?) and hooks that call check_change and stop your agent before it removes or duplicates something protected. One command per agent.
Hooks and check_change are on Solo and Team, with a 14-day trial on Free if you build alone. The deterministic MCP is on every plan.
claude › Delete src/routes/export.ts
docdrift ✖ Blocked. This removes http:GET /api/export, a protected
entry point of Export to CSV. Restore it, or mark it
"Removed on purpose" in Docdrift.
AI builders with custom MCP connect too.
For your developer
Developers get less noise, not more.
- Clear requests with a what, a why and at most five criteria they accept or edit.
- A technical annex with verified files and symbols, ready for their agent.
- One comment per pull request, only when something is high-confidence.
- No individual metrics. Nobody tells them how to build it.
Questions
Questions teams ask.
Our developers or agency won't want to install anything.
The base app only reads, publishes a check and leaves one comment. It can't push code or merge, and it doesn't measure people.
Will it give us estimates in hours?
No. It gives a relative size and the reason, next to your developer's size.
Does it replace a code review?
No. It doesn't review bugs, style or security. It answers "does this match what was agreed?"
Does everyone need GitHub?
No. One person connects GitHub; everyone else signs in with their email.
Security
Your code stays yours.
- We never store your source code.
Read in memory during analysis and deleted when the job ends.
- Analysis runs in the EU (Google Cloud).
Zero data retention. Never used to train models.
- Least-privilege GitHub access.
The base app can't push code, merge or create issues.
- Nothing changes without a person.
- Private by default. No per-person metrics.
Pricing
Start free. Pay when you depend on it.
Team is one price for you and up to 10 members, with no per-seat price.
Team
For founders who work with developers or an agency
$99a month
Yearly: $79/mo · $948 total · save 20%
- You + up to 10 members, no per-seat price
- Up to 300,000 lines of code across all projects
- 1,000 Checks a month
- 100 Change Requests or Delivery Checks a month
- 300 Ask questions a month
- Agent hooks and pre-write checks
- History: Unlimited
No card needed. Your 14 days start when your first repository is ready.
Every plan includes export, Docdrift Verified, Claude and ChatGPT, and the MCP.
No surprise charges. Extra packs are off unless you turn them on, and always have a limit you set. When you hit your quota the check shows Quota reached; nothing fails silently. Cancel anytime in Settings. Partial periods aren't refunded. The Team trial never charges you unless you add a payment method.
Ask your first question about your product.
14-day Team trial. No card needed.
Docdrift · 1 finding
Removed · Export to CSV lost 1 of 2 entry points (
http:GET /api/export).Not mentioned in this pull request or its request.
Open in Docdrift