Cursor, Devin, Claude Code, Copilot ship code 10× faster. They break things 10× faster too, because they can't see what they're touching. CodeRadius is the architecture layer your agents read over MCP, so the first diff is the right one.
Don't want to click the terminal? View the demo dashboard →
Your engineers, your AI agents, and your CI all need to know the same thing: what services exist, how they talk, and what breaks when something changes. We build that map once and hand it to each in the format they can actually consume.
Cursor, Devin, Claude Code, Copilot Workspace. They all speak MCP. Point them at CodeRadius and they stop guessing. They see who owns what, what depends on what, and what a refactor is about to touch.
Run one command and you get an HTML file with your entire system inside it. Drop it in Slack. Email it to Monday's new hire. Open it in any browser, no server to run, nothing to install, no login.
Block PRs that break architectural invariants. Doesn't matter if the author is a human or an agent. Touch a tier-0 contract without owner sign-off and the build fails. Better than finding out at 3am.
A new hire opens the architecture file you sent in their welcome email. They see what talks to what before they touch a line of code. That whole "so what's in our codebase, exactly?" conversation takes ten minutes, not two weeks.
An agent on Cursor gets a refactor ticket. Before writing the diff, it asks the MCP server what the change actually touches. The PR comes in informed. Reviewers stop playing detective with diffs.
That same PR would have silently broken a tier-0 contract. CI catches it before merge. The owner gets a Slack ping. The on-call engineer goes home on time.
Three different people. Same map. We didn't build a gatekeeper, we built the thing that your seniors carry in their heads and made it available to everyone, including the agents.
We parse the source. We also parse what ties the source together: APIs, GraphQL, message topology, database schemas. Across the languages real platform teams use, not just the ones that look good in demos.
Adding more every month · see the supported list →
If you've already invested in Backstage, Cortex, OpsLevel, or just a folder of OpenAPI specs and a CODEOWNERS file, good. Keep all of it. We read those files as signal and stitch them into the graph next to what we extract from the code.
catalog-info.yaml. You don't maintain a second one.
Sit CodeRadius next to your Internal Developer Portal, not in front of it. We consolidate the signals. You keep the workflows your team already trusts.
CodeRadius runs inside the network your security team already approved. We could make context faster by centralizing your codebase on our infra. We don't. That's the wrong tradeoff for any buyer who's been through a vendor breach before.
Every refactor at a growing engineering org survives because a handful of senior engineers can hold the whole architecture in their heads. They're the reason a Tuesday migration doesn't take billing down on Friday.
I spent the last decade as one of those engineers. It feels great, but it's a terrible way to scale a company. When the map only lives in your head, you become a bottleneck, and on-call becomes a nightmare.
CodeRadius is what that role looks like as software. The map gets built every commit. The answers are there before someone has to be paged, and the AI agents your team is already using finally get the context they were missing.
Eight engineering leaders are joining as design partners through Q3. If your stack has outgrown the people holding it together, let's talk.
Install the CLI, point it at any repo. The graph is built on your machine, in your terminal. Polyglot stacks supported. Nothing leaves your laptop. No signup, no waiting list.
This is after the MCP launch, not in it. We're telling you about it now so you can plan against it.
Take everything you've declared about your architecture: Backstage catalogs, OpenAPI specs, CODEOWNERS. Compare it to what your code actually does. Where they disagree, that's drift. You get a score per service, a list of specific gaps with file pointers, and a path to close them.
Honest answers. Including the things we don't do yet.
Those are developer portals. They ask you to maintain a service catalog in YAML. That catalog drifts the second your team ships faster than they document, which is always.
We build the graph directly from your code, every commit. The model is derived, not declared. Sit us next to your IDP. We kill the manual catalog work, not the portal.
Sourcegraph indexes code for search and navigation. We model architecture: services, contracts, dependencies, ownership, blast radius. Different problem, different data shape.
If your question is "where is this symbol used?", use code search. If it's "what breaks when I change this contract?", you want a graph.
No. CodeRadius runs locally or in your CI. The graph is built in-place. We don't store or upload your source.
LLM calls go through your own API keys (Anthropic, OpenAI, Bedrock, Vertex, Azure OpenAI). We don't proxy your prompts.
Today: TypeScript / JavaScript, Go, Python, PHP. Java and Rust come next.
On top of the source, we parse what ties services together: OpenAPI, GraphQL, gRPC, Kafka, and database schemas (Postgres, MySQL, MongoDB). Got something we don't cover? Tell us on the call. We prioritize what design partners actually need.
A 200k-LOC polyglot monorepo: full graph in under 5 minutes. Incremental (only changed files): under 30 seconds. cr blast in CI won't add real latency to your PR pipeline.
No. We don't run your code. We model your architecture.
What we do is make your tests more targeted. When a PR lands, we tell you which integration tests cover the actual blast radius. Stops you from running the full 90-minute suite for a one-line change to a tier-2 service.
We're onboarding eight engineering teams as design partners through Q3. The first call starts with what you're trying to ship, not with a price list.
Partners get founder-level access, a private roadmap line, and a say on the v1 protocol surface. We ask for weekly product feedback and the option to call you as a reference later. That's it.
Eight engineering teams onboard as design partners through Q3. We review every application personally and reply within a week. This is not a mailing list.
Prefer to talk first? Email Emanuele directly.