Built from code
Manifests, source and IaC, not a catalog someone maintains.
RigelProductMap
Rigel Map
See the whole constellation of services.
A living chart of the services, APIs, events and dependencies across your repos. Built from code, not hand-maintained, and current on every merge.
✓ Node.js / TypeScript service discovery today · Java, Python and Go Coming soon
A sample estate, charted
Four repos. One constellation.
Illustration. The repos orders-api, fulfillment, billing and notifications are read from code and drawn as one constellation of services linked by HTTP calls, events and a shared package. A pull request merges to main; only the orders service is re-analysed, two of its seven analysis passes re-run, and one rolling docs pull request is updated.
7 passes per service
Identity to integration manifest.
Code graph
AST-level, no LLM involved.
1 docs PR
Per branch, refreshed on every merge.
The rolling docs PR updates. One PR per branch, merged when green.
Deep Survey
Chart a codebase in one run.
Point Rigel at a repo. Deep Survey, a durable, checkpointed pipeline, does the archaeology and hands it back as a pull request.
1
Clone
2
Discover services
3
Scan existing docs
4
Build code graph
AST · no LLM
5
Map dependencies
6
Extract IaC
Kubernetes, AWS templates
7
Seven analysis passes
per service
8
OpenAPI + README
9
Open docs PR
Seven analysis passes, per service
Service identity
What it is, what it owns, why it exists.
API and data contracts
Endpoints, payloads, schemas.
Message and event contracts
What it publishes and consumes.
Business logic
The rules the code actually enforces.
Config and infrastructure
Env, IaC, runtime shape.
Failure modes
What breaks, and how.
Integration manifest
Every connection in and out.
- Checkpointed. Every step is named and retryable, so a restart picks up where it left off.
- Live progress in the UI while the survey runs.
- Stored per repo and branch, embedded for semantic search, opened as a docs PR.
- Fails fast on very large repos with a clear message, instead of running for hours.
Day one on an inherited estate
- Onboard 12 repos and group them into a constellation.
- Deep Survey opens a docs PR per repo.
- Constellation assembly writes the integration graph and data-flow narratives.
- Ask: "What breaks if the billing service is down?"
- Add Rigel MCP to your coding agents.
Start a Deep Survey · API
# Start a Deep Survey on your Rigel server curl -X POST https://rigel.your-company.internal/api/onboard \ -H "Authorization: Bearer $TOKEN" \ -H "Content-Type: application/json" \ -d '{"repoUrl":"https://github.com/your-org/orders-service","branch":"main"}'
The server hostname is a placeholder for your own deployment.
Quick local scan · CLI
# Quick local scan, not a Deep Survey rigel analyze --path ./orders-service --llm
The rigel CLI name is shown for the upcoming release.
Code graph
Every caller, before you touch a line.
An AST-level graph of functions, classes, files and imports, clustered into communities. No LLM involved.
AST · no LLM Optional component of your deployment
Callers and callees
For any function, including code nobody remembers.
Imports and class hierarchy
Across the whole repo.
Community clusters
How the code actually groups, not how the folders say it does.
N-hop subgraphs
Around anything you're about to change.
In the UI and over MCP
An interactive view in Rigel, plus seven code-graph tools for your agents.
reserveStock()
- Callers — 3 callers of
reserveStock(), across two clusters. placeOrder(), retryCheckout(), CartService - Imports —
reserveStock()imports from the inventory and persistence clusters. stock-ledger.ts, db/client.ts - 2-hop — The 2-hop neighbourhood: everything within two edges of
reserveStock(). Warehouse, migrations/, api/routes.ts
Clusters · CHECKOUT · INVENTORY · PERSISTENCE
Constellations
Fifteen repos. Rigel treats them as one.
Group related repos into a constellation (an Application in the UI). Rigel reads every member's deep analysis and writes the cross-service picture.
Integration graph
A synthesised account of how the services in the constellation connect.
Data-flow narratives
How data moves through the system, end to end, in plain language.
Scoped everything
Chat, bug triage, release analysis, solution design and knowledge-base learnings all work at constellation scope.
A visual cross-repo constellation map Coming soon
Today the cross-repo picture arrives as assembly documents and through the cross-repo tools over MCP.
Already have a developer portal? Keep it.
Portal catalogs are maintained by hand. Rigel's chart is derived from code, so use it to check which services exist, what they expose and who depends on whom.
Catalog sync and drift reports are on our radar. Tell us which portal you run.
Living Docs
Docs that update themselves on every merge.
No stale wiki and no fifty doc PRs. Rigel re-analyses only what the diff touched, and delivers it as one reviewable PR.
Merge a PR to main · Illustration
| ID | API | Events | Logic | Config | Failure | Integr. | |
|---|---|---|---|---|---|---|---|
| gateway | |||||||
| orders | ↻ | ↻ | |||||
| fulfillment | |||||||
| billing |
Triage: 2 of 28 cells. orders × API contracts and configuration. Nothing else re-runs.
Triage: 2 of 28 cells. billing × event contracts and integration manifest. Nothing else re-runs.
Triage: skipped, tests only. No analysis pass re-runs and nothing is committed.
docs: refresh living docs
gateway/service-identity.mdearlierorders/api-data-contracts.md+ updatedorders/configuration-infrastructure.md+ updated
3 commits · 1 open PR · merges when green
Refresh knowledge base on merge On
default: on
Generate OpenAPI spec on merge Off
default: off · 14 endpoints detected
1
A PR merges
to a branch you track.
2
Triage reads the diff
and picks the service-by-pass cells it affected.
3
Only those cells re-run
Docs-only, tests-only and lockfile churn are skipped.
4
One rolling PR updates
All changes for a branch land in a single PR.
5
OpenAPI can follow
Turn it on per repo; a valid spec is never replaced by an invalid one.
How it works
Triage is an agent instructed to skip diffs that don't change what the docs describe, and to refresh when in doubt. The rolling docs PR heals its own conflicts and auto-merges when green. Postgres advisory locks stop concurrent writers from overwriting each other.
Each refresh runs on a budget, with a USD cap per run. Docs PRs contain only generated doc paths, so in repos with no CI they can merge without a CI signal; install Repo Guard to enforce the path rule.
Docs that never drift: a PR merges to main. Rigel re-runs two passes for one service, updates the rolling docs PR and the OpenAPI spec, and the PR merges when green. The Documentation Drift Check, on the schedule you set, confirms no undocumented endpoints remain.
A chart for every branch you ship from.
Your docs describe main. Production runs release/2026.10. Rigel keeps a snapshot for each, so analysis, chat and PR impact are pinned to the branch they concern.
Merge refresh runs only on branches you track. Older snapshots are pruned automatically.
Knowledge base
A memory that flags its own contradictions.
Agents shouldn't relearn your domain language every session.
Semantic search
Every analysis document is embedded, so agents and people find the right context by meaning, not keyword.
⌕ where do we decide an order can be cancelled?
- orders · business logic — Cancellation rules by order state
- fulfillment · failure modes — What happens to a shipment mid-cancel
- orders · API contracts — The cancel endpoint and its payload
Illustration. Results match by meaning, not by keyword.
Self-learning knowledge Beta
Glossary
The terms your teams use.
Learnings
Durable engineering heuristics.
Memory
Facts from runs, chats and agent conversations.
Every write passes a fail-closed leakage critic. Nightly consolidation merges near-duplicates and marks contradictions for an admin, instead of quietly picking one.
Belief · learningsRefunds settle through the
payments.settled event.Belief · memoryRefunds settle in the nightly billing batch.
contested
Illustration. The self-learning knowledge base is in Beta and surfaced through an admin dashboard today.
Ask Rigel
Ask your architecture. Get a cited answer.
Chat with one repo and branch, or with a whole constellation.
- A tool-using agent searches services, dependencies, the API surface, communication channels and DevOps context.
- An investigation timeline shows every tool call it made to reach the answer.
- Token use and estimated cost on every session, and every query runs under a USD cap.
- Long sessions are summarised to stay within budget.
Which services publish to payments.settled?
Where is the retry policy for outbound webhooks configured?
From your terminal
rigel query "Which services call the inventory reservation endpoint?"
Ask about a local repo. Needs your Anthropic key, or Claude through your AWS account. The rigel CLI name is shown for the upcoming release.
What breaks if the billing service is down?
- ✓ searched services — billing-worker
- ✓ read dependency graph — 3 edges
- ✓ read communication channels — 2 queues
- ✓ read failure-modes doc — billing
Checkout keeps working; settlement stalls. orders publishes to the billing queue and doesn't wait on it. ledger stops receiving entries, and notifier holds receipts until billing recovers.
billing · failure modes·orders · event contracts·ledger · integration manifest
every query reports its tokens and estimated cost · under a per-query USD cap
Next
Put the chart to work.
Next in the platform →
Rigel Impact: a readable risk score for every PR, computed on the chart.
What ships today →Shipped, Beta and Coming soon, feature by feature.
Book a technical walkthrough →We'll chart a demo estate with you, then one of your repos.
Mechanism: which parts are deterministic and which use a model → · Your coding agents: 20 read-only tools over MCP →
Chart your first repo.
Node.js and TypeScript services today. Java, Python and Go discovery is coming soon.