Get started

Code intelligence, not grep

Code intelligence your agents cite, not grep.

RepoBrain indexes your repositories, any git repo, local or remote, and gives agents structure-first retrieval. A query returns the right lines of code with a file:line citation, not a file to read end to end.

RepoBrain · your code, mapped

indexed filesimport edges

In plain words

What it is, where it lives, when to reach for it

What is it
A code search index for your repositories that returns cited file and line snippets, not whole files.
Where do I use it
In the RepoBrain view of the dashboard, and from your terminal through the quox CLI.
When would I use it
When you or an agent needs to find where something lives in a codebase before reading files.
How do I use it
Run quox nucleus query "<question>" in your terminal and read the cited file and line snippets.

Not a search box, a map

Not a search box. A map of how your code connects.

Grep finds strings. Vector search alone finds nearby text. Neither knows that a function is a definition, or that two files are connected by an import. RepoBrain indexes the structure up front, so a query returns the right code, not everything that happens to match.

Grep

Blind

Matches text, not meaning. It cannot tell a definition from a comment, so it returns whole files for an agent to re-read from scratch.

Vector search alone

Guesses

Everything gets embedded and nothing gets weighed. Retrieval returns whatever is nearby, missing exact symbol names and import edges.

RepoBrain

Indexes and connects

FTS5, a symbol index, and the import graph, together. It knows what things are and how they connect, so a query returns the right lines of code with a file:line citation, not a file to read end to end.

One index, four steps

Index any repo. Query it from anywhere.

Connect a repo once. RepoBrain builds the index in a single pass, Nucleus is the query face every agent calls, and the answer comes back grounded in your actual code.

01 · Connect

Any repo

Local or remote, any git repo, picked from GitHub or pointed at a path.

02 · Index

RepoBrain

Builds FTS5, a symbol index, and the import graph, all in one pass.

03 · Query

Nucleus

The query face: `repobrain query "..."` from a CLI, or the governed API.

04 · Cite

Agent

Gets the right lines of code, cited by file and line.

Index once, query as many times as you like, from a terminal, a script, or another agent.

Grounded, not guessed

Agents cite code with file and line, not guesses.

A structure-first query does not paraphrase an answer from memory. It returns real snippets, each tagged with the file, the line range, and the repo it came from, so a human or another agent can check the claim in seconds.

An agent that greps burns your context window on file dumps and then guesses; you pay for the tokens either way. Under the hood the retrieval is structure-first: FTS5, a symbol index, and the import graph.

File:line citations, every result50-line snippets, not whole filesDefinitions, not just text matchesImport edges, not isolated hitsOrg-scoped resultsProvenanced snippets, repo and commit aware

The CLI is `repobrain query "..."`, targeting a single repo by default, or any repo with `-t <repo>`.

Explore the import graph

It sees the shape of your codebase.

This is demonstration data, not a live repo, but the shape is the same one you would see for real: files gathering by directory, hub files standing out as the ones most imported, and edges forming wherever an import actually connects two files.

ComponentsServicesRoutesUtilsTests
0 files5 directories0 imports

Structure beats brute force

The right code, not a file dump.

A structure-first query returns the right lines of code, not a file to read end to end. Agents reach RepoBrain the same way every time: one shared tool, org-scoped, with the file and line attached to the answer.

One shared tool, every agent, no whole-file dumps burning the context budget.

Y

You

Where do we validate the webhook signature?

querying RepoBrain

Q

Agent

Signature validation happens in the webhook handler, before the payload is parsed, using a constant-time HMAC comparison against the stored secret.

RepoBrainserver/webhooks.js:142quox-dashboard, indexed 6 minutes ago

tool repobrain_query("webhook signature validation")

The real interface

Query it from a CLI, or a governed API.

Not a mockup. Real output from two `repobrain query` runs against a live Quox codebase, captured tonight.

repobrain query

$ repobrain query "where is the AEE envelope schema defined"

src/lib/aee/envelope.js:41-90 (Definition of AEE)

src/lib/aee/envelope.js:361-410 (Matches: AEE, envelope)

src/lib/aee/envelope.js:161-210 (Matches: AEE, envelope)

$ repobrain query "what imports agentRegistry"

src/config/agentRegistry.js:1-50 (Definition of agentRegistry)

src/lib/aee/capabilities.js:1-50 (Definition of agentRegistry)

src/test/agentRegistry.test.js:1-50 (Contains agentRegistry)

Two real queries, six results, each an exact file and line range: no file to open and read end to end.

The difference that matters

Governed, org-scoped, provenanced.

A grep alias or a local index does not know who is asking or what they are allowed to see. RepoBrain does. Every query stays inside your organisation's connected repos, every result carries where it came from, and every request is logged as evidence, not just claimed.

Scoped

Results are limited to the repos your organisation has connected. Nobody queries across a scope they are not in.

Provenanced

Every snippet returned carries the repo, file, and exact line range it came from.

Licence-gated and audited

Access to the query API is gated per organisation, and every query is logged as governed evidence.

Watch it fill in

From first index to full graph.

Connect a repo and RepoBrain moves through the same stages every time, from a first pass to agents citing it every day, and finally into Brain2's own graph.

Stage 1

Connected

A repo is linked, local or remote, not yet indexed.

Stage 2

Indexed

FTS5, the symbol index, and the import graph are built in one pass.

Stage 3

Queried

Agents are citing file:line answers from it in everyday work.

Stage 4

Bridged

The index feeds Brain2: files as neurons, imports as axons.

Explore the import graph node by node above, or query it directly with `repobrain query "..."`.

What is live today

Live now, and in progress.

Live now

  • Structure-first retrieval: FTS5, the symbol index, and the import graph
  • The `repobrain query` CLI, with `-t <repo>` and cross-repo `-x a,b`
  • Connect any repo, local or remote, including a GitHub picker
  • The governed per-org API, scoped, provenanced, and audited
  • The import graph explorer
  • The Brain2 bridge: your codebase as neurons and axons in the living graph

In progress

  • Deeper cross-repo ranking for the `-x` cross-repo query mode
  • Broader language coverage for the symbol index

Straight answer: a cold query took 42 to 54 seconds against this instance tonight, and 22 seconds against the newer v2 host. Indexing is per target, not instant across every connected repo at once, and RepoBrain runs as a separate host from the agent asking the question, one that can be down: it was, earlier this evening.

Two brains

RepoBrain reads code. Brain2 remembers everything.

One plugin, two names, and a peer product it feeds. Here is what each part actually does, and where the lines are.

RepoBrain · this page

The code intelligence engine

Indexes any repo, local or remote, into FTS5, a symbol index, and the import graph. This is the product you enable and connect repos to.

Nucleus

The query face

What agents actually call: the `repobrain query "..."` CLI, or the governed per-org API. Same index, two ways in.

Brain2

The memory layer

Remembers everything else: conversations, docs, team logs, evidence. Your codebase feeds into it as neurons and axons alongside everything it already knows.

See Brain2 →

How they fit: RepoBrain indexes your code and exposes Nucleus as the query face every agent calls, and the same index feeds Brain2, so your codebase shows up in the living graph as neurons and axons alongside everything else the platform knows.

The answer, cited

Ask where something lives, get file and line

repobrain query3 results in 1344ms
$ repobrain query "where is the skill pack installer"
  1. services/collector/lib/skillPackInstaller.js:237-241score 1.46
    function resolvePackForCapabilities(enabledCapabilities) {
      const caps = sanitizeCapabilities(enabledCapabilities);
      if (!caps.includes('dev.register')) return null;
      return CORE_DEV_PACK;
    }
  2. services/collector/lib/skillPackInstaller.js:406-409score 1.46
    function buildStatusCheckCommand(pack) {
      const base = remoteBaseDir(pack.subdir);
      return `cat "$HOME/${base}/${PACK_VERSION_MARKER_FILE}" ...`;
    }
  3. services/collector/lib/skillPackInstaller.js:427-428score 1.46
    async function fetchPackVerification(packId, orgId) {
      const url = `${QUOXFLOW_URL}/api/skill-packs/${packId}/verify`;
Targeted, cited snippets, not a 10,000 token file dump. Your agents cite code by file and line, the way a reviewer would.

How you get it

In the store today, and in the bundle

store

On its own

RepoBrain is the founding Dev Suite member and the proof the model works: a live store listing paid for by real customers, installed on your own QuoxCORE instance with the quox CLI verbs included.

$39 · one-time · in the store
bundle

In the Dev Suite

RepoBrain also belongs to the Dev Suite, alongside the plan, the loops, the verdict and the rest of the toolbox. The suite does not have its own store listing yet, so today you buy the members you want individually.

no separate listing yet

Give your agents code they can cite.

Connect a repo, index it once, and query it from anywhere. Every agent gets targeted snippets with file and line, not a whole file to read end to end.