Turn a GitHub issue into a verified pull request
Torque orchestrates an investigation via Entire, creates isolated test worktrees, runs your test suite, and authors a pull request. If tests fail, a Recovery agent analyzes neatlogs telemetry and retries with a new strategy.
Run Torque Workspace
Supply a GitHub issue to simulate the multi-agent investigation, execution, and recovery loop.
How it works
A transparent, stage-gated pipeline from issue diagnosis to peer-reviewed PR.
Connect GitHub
Authorize Torque on your repository with fine-grained read and pull request access.
Select Issue
Point Torque to a specific GitHub issue or bug ticket from your authorized repository.
Investigate & Implement
Investigator queries Entire for call sites; Implementer applies code edits and runs tests in an isolated worktree.
Recover If Needed
If initial tests fail, Recovery reads neatlogs trace spans, isolates the flaw, and directs a new strategy.
Review Verified PR
Verifier audits acceptance criteria and opens a pull request with complete execution evidence attached.
Getting started
Connect your account, choose a repository, and let Torque handle the rest. Access permissions are kept minimal and strictly scoped.
Connect an account
Log in through GitHub OAuth to identify your developer session.
Authorize a repository
Install the Torque app exclusively on the repositories where you want fixes created.
Supply or select an issue
Paste the URL of an issue or bug report requiring remediation.
Run Torque in the workspace
Monitor live agent attempts, trace logs, and inspect the resulting pull request.
Plain-Language Access Requirements
Torque requires repository read access to analyze your code and write access to push patch branches and open PRs. Torque does not:
- Merge code directly to main or protected branches
- Access repositories outside your authorized list
- Store private source code beyond active execution jobs
Why Torque
Built on three foundational principles for trustworthy code generation.
Test-gated changes
No patch is proposed without verification. Torque runs your project's existing tests and newly synthesized regression tests in clean worktrees before opening a PR.
Bounded recovery
When fixes fail, Torque doesn't hallucinate repeatedly. The Recovery agent inspects telemetry traces to alter strategy with explicit retry caps to avoid infinite loops.
Inspectable evidence
Every decision, diff, pytest run, and recovery rationale is recorded in neatlogs trace spans and Entire checkpoints for full human audibility.
Evidence & Integrations
Torque integrates with production-grade tooling created during neatHack.
neatlogs
Execution Telemetry & Traces
Provides high-fidelity trace logging of agent tool calls, execution times, and pytest assertion failures that feed directly into the Recovery agent's diagnostics.
Entire
Codebase Context & Checkpoints
Powers the Investigator's code retrieval, call graph navigation, and checkpoint state snapshots to ensure patches target the correct architectural layers.
cfo.ai
Business & Unit Economics Plan
The business strategy, unit costs, and ROI models for autonomous bug resolution and developer acceleration created for the neatHack submission.