Self-hosted · Source-available · BYO model

Grok for your own product .

Ask anything about your code, infra, tickets, and tests — with citations. Self-hosted, so the answers and the sources stay on your servers.

Want pricing, a demo, or to talk through your stack? Drop your email — we'll reach out.

#engineering · team chat
Slobodan 14:32

@Oracus how does our abandoned cart recovery work?

Oracus.ai bot 14:32

Ask anything

Bug coverage, release status, spec drift, on-call context, who-owns-what — if it lives in something Oracus.ai has ingested, you can ask it. Every answer cites the ticket, test, or file it came from.

Is the bug SHOP-203 (cart cleared on retry) covered by tests?

Partial — referenced by retry test [shop.checkout.retry], but it's currently failing. Scenario is documented, not verified.

What does checkout do, and what's actually tested?

Spec covers Stripe + 3DS2 [SHOP-50] and cart-on-retry [SHOP-203]. Tests verify the Stripe happy path; retry behavior is not passing.

Are we behind on the Q3 mobile epic?

14 stories in MOBILE-Q3: 9 Done, 3 In Progress, 2 Backlog. The 3 in-flight tickets all block on PLAT-118.

Which API endpoints don't have integration tests?

12 of 47 endpoints have no test referencing their handler. Highest ticket activity: /webhooks/stripe, /admin/users/:id, /reports/export.

Who shipped the rate-limiter, and where's the spec?

PLAT-91 (closed 2026-02-14, owned by @ana). Spec lives in `specs/rate-limit.md`; tests in `apps/api/tests/rate-limit.test.ts`.

What's broken on main right now?

3 failing suites in the latest run: shop.checkout.retry, billing.invoice.pdf, auth.sso.okta. All three first failed in the last 7 days.

Map your architecture, not your guesses

Oracus.ai reads what's actually deployed — Kubernetes manifests, Dockerfiles, .proto packages, OpenAPI specs, and package manifests — to derive a cross-repo service map directly from infrastructure-as-code. No diagrams to maintain, no whiteboard kept up to date by hand. The map below is generated from a real Oracus.ai ingest of a public reference architecture.

loadgenerator python · k8s frontend go · k8s adservice java · k8s shoppingassistantservice python · k8s recommendationservice python · k8s checkoutservice go · k8s productcatalogservice go · k8s cartservice csharp · k8s currencyservice nodejs · k8s shippingservice go · k8s paymentservice nodejs · k8s emailservice python · k8s
http call · go python node java c#

Covers REST, gRPC, queues, and shared libraries — refreshed on every ingest, so the map is never the stale one on someone's laptop.

Sources flow in, cited answers flow out

One grounded answer surface across every system your team treats as source-of-truth. Tickets, tests, specs, and docs — connected, embedded, and citing the source on every claim.

Git repos Jira JUnit Files Confluence OpenAPI GitHub PRs Oracus.ai embed · link · ground Cited answers

How it works

  1. 1. Connect

    Pull in your sources

    Start from your git repositories — Oracus.ai clones them and auto-extracts specs and dependency maps. Layer on Jira tickets, JUnit results, Confluence pages, OpenAPI specs, GitHub PRs, and uploaded docs. Bring your own connector via the ingest API.

  2. 2. Embed

    Index and link

    pgvector + your provider of choice (OpenAI, Anthropic, Google, Ollama). Sources get cross-linked so a question about a ticket surfaces the matching test, spec, and doc.

  3. 3. Ask

    Grounded, cited

    Every claim cites a source id. The prompt enforces it; the response schema requires it. Hallucinations are structurally hard.

Know who shipped what

Connect your GitHub org alongside Jira and Oracus.ai stitches them together. Every ticket gets resolved to the engineers and teams that worked on it — and every answer can name them.

People, not generic owners

"PLAT-91 was shipped by @ana on the Payments team" beats "the engineering team owns this." Oracus.ai pulls org members, team memberships, and PR authorship and joins them against Jira assignees and reporters.

Onboard new engineers in hours

"What does the Payments team work on, and who runs it?" returns recent ticket history, current in-flight work, and a named tech lead — straight from Slack, no onboarding doc required.

Trust the ownership signal

A claim that shows up in both systems — same person assigned the ticket and authored the PR — beats either signal alone. Cited owners come from the intersection, not a single weak source.

@Oracus.ai who can I ask about our 3DS payment flow?

@miguel built it in SHOP-50 (Payments team). There are 2 open follow-up tickets owned by @ana, and the test suite around it is currently passing.

Read-only GitHub access is enough — Oracus.ai only needs the org's member, team, and PR metadata to build the ownership graph. No code is ingested through this integration; that comes from the git repository connector.

Read the GitHub integration docs

Why now

Engineering orgs have grown beyond the point where one person remembers everything. Tickets, tests, specs, and docs rot independently. Oracus.ai is opinionated about grounding so the answers stay honest, even when the corpus is messy and contradictory.

Self-host first

One Docker image, one Postgres, your API keys. Sensitive context never leaves your network.

Provider-agnostic

Vercel AI SDK underneath. Swap models per task. Bring your own embeddings.

Citations, always

The model never gets to skip evidence. The prompt enforces it; the schema requires it.

Inspect what you run

Source-available. Your engineers can audit every line that runs inside your network — no black-box agents on your infra.

Talk to sales

Pricing scaled to your repo count, a guided demo, or a chat about how Oracus.ai would fit your stack. Leave your email and we'll be in touch.