Security and data handling.

What happens to a repository you connect, the controls in place today, who else touches your data, and what we have not done yet.

Accountable Paramveer Singh, founder. Questions a security review needs answered go to [email protected].

Where we stand on SOC 2

We have not completed a SOC 2 audit. We are preparing for a Type I report against the Security criteria, to be followed by a Type II observation window. The controls below are the ones in place today.

The product also assesses your readiness against 126 SOC 2 and ISO/IEC 27001:2022 controls. That is a readiness assessment, not a certification of you or of us.

Controls in place

Isolation
Each workspace’s data is isolated by Postgres row-level security, forced for the application’s database roles. A project is a scope inside a workspace, never a boundary between customers.
Sign-in
Every user needs a second factor. Passwords are hashed with scrypt; session tokens and API key secrets are stored only as hashes.
Credentials
Tokens and keys you connect are sealed with AES-256-GCM. Cloud accounts are connected by a read-only grant you make in the provider’s own console.
Your code
Analysed as data, never executed. Scanners run in a container with no network, under a user of their own, on a copy they cannot write.
Transport
TLS 1.2 or later with HSTS on every public host, and a strict content security policy.
Backups
Daily, encrypted before they leave the server to a key whose private half is kept off it, and kept 30 days off-site.
Operators
The operator console is reachable only from allowlisted addresses and needs a second factor. Every operator action is written to a hash-chained audit log.
Audit trail
A workspace audit log records sign-ins, membership and role changes, connectors, AI settings and review runs, and a workspace admin can read it.
Change control
Every change goes through a pull request and automated gates (lint, type checks, tests, a secret scan) before a separate release step.

Models and your code

The findings come from deterministic tools and our own analyzers. A model explains, summarises and drafts.

Default
AI analysis is on for a new workspace. An admin can switch it off for the workspace or one project, and every review records which way it was set.
When on
Code excerpts and finding text go to the model provider the job uses, and code chunks go to OpenAI for search embeddings, with email addresses and card numbers masked first.
When off
Your code goes nowhere but our server and the services you connect yourself.
Training
Never. No model is trained on your code, your findings or your reports.

What we keep, and for how long

Cloned trees
Deleted on a schedule after the review that needed them.
Archives
Virus-scanned before anything reads them, and deleted when you ask or when their retention ends. Only the name, digest and verdict are kept.
Findings
Kept for the ten most recent runs of each kind in a project.
Account data
Names, email addresses and sign-in records, kept while the workspace exists.

Subprocessors

Providers you connect yourself, such as Slack or a cloud account, receive only what you send them.

Provider Purpose Receives
Hostinger Hosts the product and its database in Mumbai, and holds its encrypted backups. Everything the product stores; backups only in encrypted form.
Anthropic Language models, only while AI analysis is on. Code excerpts and finding text, masked.
OpenAI Language models and search embeddings, only while AI analysis is on. Code excerpts, finding text and code chunks, masked.
Google Language models, only while AI analysis is on and a job is routed to them. Code excerpts and finding text, masked.
GitHub The source-control integration you install. Repository contents it already holds; the comments and check runs we post.
Amazon Web Services Account email (SES). Recipient addresses and the links in confirmation, invitation and reset messages.
Cloudflare DNS for every host, and hosting for this site. No product data: the app and its API are not proxied through it.

Not done yet

  • No SOC 2 report yet, and no external penetration test.
  • No SAML or SCIM yet. Sign-in with Google or Microsoft Entra is available.
  • One server, so no high availability; a restore has not yet been rehearsed and timed.
  • Deleting a whole workspace is done on request, by hand.

Report a vulnerability

Write to [email protected] with what you found and the steps to reproduce it. Test only against a workspace you created yourself, never another customer’s, and do not run load or denial-of-service tests. We do not offer a bounty.

The same address is published in security.txt.