Every PR review is a decision.
Your agent should remember them all.
Connect GitHub. Hebbrix turns your repos, issues, and pull requests into memories your AI agent can reason about.
The context nobody reads
Six months ago someone on your team opened an issue explaining why you can't just “swap out the message queue.” It has four paragraphs of context about latency constraints, two diagrams embedded as images, and a reply from your CTO saying “agreed, let's keep RabbitMQ.” That issue is the canonical source of truth for that decision. It has 0 upvotes and no one will ever search for it.
Meanwhile, a new engineer is about to propose switching to Kafka in their next PR. They'll spend a week on the migration plan before someone mentions the old thread in standup.
This is the kind of thing Hebbrix fixes. Not by building a search engine on top of GitHub (you already have that). By turning those scattered threads into structured memories that your AI agent retrieves automatically when they're relevant.
After you connect
This is what conversations with your agent actually look like once GitHub is synced.
Based on issue #189 in acme/platform-core, the team evaluated Kafka in Q2 but decided against it. The main concerns were tail latency under your current partition scheme and the operational overhead of managing a Kafka cluster for your message volume. Your CTO approved keeping RabbitMQ with the caveat of revisiting if throughput exceeds 50k msg/s.
acme/billing-service has 3 open PRs. @sarah opened #254 (fix race condition in concurrent subscription updates, 2 approvals). @marcus has #251 (migrate invoice PDF generation to async workers, merged yesterday) and #258 (add retry backoff for failed webhook deliveries, in review).
There are 4 open issues on acme/payment-gateway. The critical one is #112: idempotency keys aren't enforced on the /charge endpoint, which means retried requests can double-charge. @sarah flagged it two weeks ago. Also worth noting: #98 tracks PCI compliance changes due by end of quarter.
Sync scope
Read-only. Hebbrix never writes to your repos.
README, description, languages, topics, visibility. Your own repos are weighted highest.
Descriptions, review comments, merge status, branch info. Linked to repos and contributors.
Full body and comments, labels, assignees, state. Connected to parent repositories.
The libraries and tools you bookmarked. Synced as reference material with lower weight.
The technical details you care about
Authentication: Standard OAuth 2.0 flow. Hebbrix requests repo and read:user scopes. You choose which repositories to include. Revoke access from your GitHub settings at any time.
Access: Read-only. Hebbrix never pushes commits, opens issues, or modifies your repositories in any way.
Sync cadence: Initial sync runs immediately after connection. After that, incremental sync every 30 minutes picks up new and updated items only. No full re-scrape.
Deduplication: Every synced item gets a source_reference tied to its GitHub URL. If an issue is updated, the existing memory is patched, not duplicated.
Give your agent a codebase-sized memory
Free tier. OAuth connect. Under a minute to set up.