AI development platform
Your team’s agent setup, run against your repos
Torque keeps your skills, hooks, CLAUDE.md and AGENTS.md files and model settings in one library. It runs coding agents against your repositories in containers, and a review agent reads every merge request before you do.
- An AI-ready ticket enters the queue.
- An agent runs in a container with your organisation’s skills, hooks and rules.
- A review agent reads the merge request and comments.
- You review and merge.
AI coding works on one laptop. It does not work across a team.
The setups that work are personal. Nothing carries them to the next person, the next repo or next month.
Every developer has their own setup
Prompts, CLAUDE.md files and habits live on one machine. A good setup stays where it was written.
No shared process
Nobody agrees which skills apply, what an agent may touch, or when a person looks at the result.
No record
When an agent changes code, the reasoning is gone with the terminal that ran it.
How it works
Connect your Git host, build one library, queue tickets, review the results.
- Built today
Step 1: Connect GitHub or GitLab
Install the Torque GitHub App or add a GitLab token. Torque reads repository and branch metadata. It does not clone or index your code.
- Coming next
Step 2: Build your library
Put skills, hooks, CLAUDE.md and AGENTS.md files and model settings in one organisation library. Apply them to projects, with locks where the organisation decides.
- Coming next
Step 3: Queue AI-ready tickets
Write tickets with the context an agent needs. They enter a queue, and agents work them in containers with your library inside.
- Coming next
Step 4: Review, then merge
Each run opens a merge request. A review agent reads it first. You decide last.
For developers
Your local setup, run the same way in the cloud.
Coming next Bring what already works
The skills, hooks and .claude/ or Codex configuration you use today run the same way inside a Torque container.
Coming next Questions go to your inbox
When an agent needs a decision it asks in your inbox and carries on with the next step. Nothing waits on you.
Coming next Tickets with full context
Repo, branch, related files and the rules that apply are attached before the agent starts.
For organisations
One library, one set of rules, inherited by every project.
Settings inherited with org locks
Built todaySettings flow from the organisation to projects and repos. Lock the ones that must not change.
Roles and restricted projects
Built todayFixed organisation roles with project-level overrides. Restricted projects are visible only to the people granted access.
A database per organisation
Built todayEach organisation’s data lives in its own Postgres database, created when the organisation is.
Claude and OpenAI, your way
Coming nextUse subscriptions or your own API keys. Keys are stored encrypted and decrypted only inside a run.
What exists today, and what comes next
Torque is in private development. This is the honest state of it.
Built today
Organisations and roles A database per organisation GitHub and GitLab connections Projects with repositories and branches
Coming next
The agent library: skills, hooks, CLAUDE.md and AGENTS.md, models AI-ready tickets Agent runs in containers AI review on merge requests Flows and routines Knowledge bases
Security
Built in from the first milestone. All of this exists today.
Built today A database per organisation
No shared tables between organisations. Isolation is at the database level, not a column.
Built today Envelope-encrypted secrets
Each secret is encrypted with its own data key. Data keys are wrapped by a master key held outside the database.
Built today Public-internet-only Git host access
Self-managed GitLab must be reachable over public https. Private and loopback addresses are refused.
Built today Verified GitHub App installs
An installation is linked to an organisation only after the installing user proves they can see it.
Questions and answers
Short answers to the questions we get asked most. Everything here is true today, and anything not yet built is marked coming next.
What is Torque?
Torque is an AI development platform for teams. It keeps your organisation’s agent setup in one library: skills, hooks, CLAUDE.md and AGENTS.md files and model settings. It runs coding agents against your repositories in containers and puts a review agent on every merge request before a person looks at it. Built today: organisations and roles, a database per organisation, GitHub and GitLab connections, and projects with repositories and branches. Agent runs and AI review are coming next.
Does Torque work with Claude and Codex?
Claude from Anthropic and Codex from OpenAI are the first two agents. Torque runs the same skills, hooks and configuration files you use with them locally, with model settings held in your organisation library so every project inherits them. This part of the product is coming next.
Do I need my own API keys, or can I use a subscription?
Either. Use a Claude or OpenAI subscription, or your own API keys. Keys are stored encrypted and decrypted only inside a run. Usage is billed by the provider to the account the key or subscription belongs to. Coming next.
Where do my code and data live?
Each organisation has its own Postgres database, created when the organisation is. In the current release Torque reads repository and branch metadata from GitHub or GitLab; it does not clone or index your code. When agent runs arrive, they will run in containers and the code they change goes back to your Git host as a merge request. The hosting region is stated in the privacy policy.
What exists today, and what comes next?
Built today: organisations and roles, a database per organisation, GitHub and GitLab connections, projects with repositories and branches, envelope-encrypted secrets and verified GitHub App installs. Coming next: the agent library, AI-ready tickets, agent runs in containers, AI review on merge requests, flows and routines, and knowledge bases. Every feature on this page carries one of those two labels.
How do I get access?
Join the waitlist with your email. We send one confirmation email, and we write again when there is something for you to try. There is no fee during the waitlist and beta. Every email carries an unsubscribe link.
Who is Torque for?
Development teams that already use coding agents and want one setup, one process and one review step across every repository rather than one developer’s laptop. Platform and DevOps engineers who need roles, restricted projects and a database per organisation will find those built today.
Join the waitlist
Leave your email and we will write when there is something to try. One confirmation email, no newsletter.