Haystack

Haystack is an AI-powered code review tool built to reduce the burden of reviewing every pull request manually. It reviews all incoming PRs, applies your team's learned standards, aggregates signals from multiple AI reviewers, and routes only the issues that genuinely need human attention to your feed. The tool is available via a web-based interface and a CLI (haystack submit).

What It Is

Haystack sits between your GitHub repositories and your engineering team, acting as an automated triage layer for pull requests. Rather than replacing code review entirely, it filters the noise — auto-merging clean PRs, surfacing real bugs, and routing sensitive changes (such as those touching auth or public APIs) to the right reviewer with full context. It combines outputs from multiple AI reviewers (including CodeRabbit, Copilot, Greptile, BugBot, and Gemini, per the homepage) into a single triage decision instead of generating multiple separate bot comments.

How the Triage Workflow Works

The core workflow follows a structured pipeline:

• Standards enforcement: Haystack scans past reviews to extract the patterns your team already enforces — test coverage thresholds, API compatibility rules, documentation requirements — and applies them consistently on every new PR.
• Pre-submit checks: Running haystack submit triggers pre-flight checks before the PR is even opened, catching issues at the earliest possible point.
• Agent context filtering: Haystack reads the author's coding-agent session checkpoints to drop false positives — findings that the author already addressed or that contradict real design decisions made during development.
• Auto-fix (Alpha): With the --auto-fix flag, mechanical issues like missing null checks or bitmask corrections can be resolved automatically in the background; ambiguous or risky changes still go to a human.
• Merge queue: PRs that pass triage enter a merge queue where Haystack handles rebases, conflict resolution, and CI repair automatically.

Continuous Learning

Haystack tunes itself over time based on reviewer feedback. Policies that get dismissed repeatedly are removed; repeated bug patterns from past reviews become new rules. The homepage illustrates this with examples such as a test-coverage threshold being relaxed for docs-only PRs after generating too many false positives, and a new concurrency rule being added after a data race was identified in a reviewed PR.

Target Audience and Use Case

Haystack is aimed at engineering teams that receive a high volume of pull requests and want to protect reviewer attention. The product's framing centers on the idea that not every PR deserves the same level of scrutiny — sensitive changes touching event loops, authentication, or public APIs get routed with full context, while routine changes merge without interruption. The CLI-first submission flow (haystack submit) suggests a developer-centric workflow integrated into existing terminal habits.

Current Status

Haystack is actively available with a 14-day free trial (no credit card required). The auto-fix feature is explicitly labeled Alpha on the homepage, indicating it is in early availability. The product is accessible via web login with GitHub authentication.