Deterministic graph analysis, blast radius, and why the model has never seen your source.
CrossCodeEval is the best-shaped benchmark for cross-file code intelligence in TypeScript. Here's what the benchmark reveals, where its methodology has gaps, and what it means for evaluating tools like Spiderbrain.
Read →Most post-mortems identify the same root cause: a high-blast-radius change with no test coverage on the callers. A code graph surfaces that combination before deploy, not after the outage.
Read →Every new AI session starts cold. No memory of the last session, no knowledge of past decisions, no awareness of constraints established three weeks ago. Here's the real cost of session amnesia, and how a persistent project memory changes the economics.
Read →Larger context windows don't make AI coding assistants more accurate, relevant context does. A code graph lets you compute exactly which files belong in the window for any given task, without trial and error.
Read →A prose prompt describes a system. A blueprint shows one. The difference is precision: relationships, types, and structure that prose leaves ambiguous are explicit in a canvas. Here's why that changes what the AI builds.
Read →Your team has services, packages, and folders. Your codebase has communities, emergent clusters of tightly coupled code that rarely match your org chart. Finding them changes how you refactor, plan, and own code.
Read →The first week on a new codebase is expensive, for the engineer and for the team answering their questions. A code graph turns that week into a morning by answering the structural questions that no amount of documentation can.
Read →One hosted MCP connection gives Claude, Cursor, Claude Code, or any MCP client access to 28 Spiderbrain tools, codebase, memory, and blueprint. No local server. No glue code. No reconfiguration per client.
Read →Leave a note on any memory in the Memory Tree. Your AI reads it on the next session, applies the update, and resolves the note. Steering your build doesn't require a new conversation, it requires annotating the memory.
Read →The AI built something. It works. But you don't fully understand the architecture it chose. Here's the audit workflow: use Deepview to read the structural story, find the decisions worth preserving or reversing, and rebuild your mental model from the graph.
Read →A blindspot file has low in-degree, it's not widely imported, but high transitive out-reach. Reviewers skip it. AI agents underrate it. Then it appears in your post-mortem. Here's how to find them before that happens.
Read →Blast radius measures how many modules a change can break. Every codebase has it. Almost no tool surfaces it. Here's why it's the most important number you're not looking at before you merge.
Read →You've inherited a large, under-documented codebase. Spiderbrain reads the actual import graph, runs community detection, and draws the architecture, not from the README, but from the code. First 5 minutes with any unfamiliar repo.
Read →Your AI assistant reads files, but your codebase is a graph, not a folder. Here's the structural blindspot that makes it hallucinate dependencies, miss blast radius, and confidently give you wrong answers.
Read →Structural drift is what happens when the codebase evolves faster than its architecture does. Deepview tracks it commit by commit, which communities shifted, which keystones gained new callers, which docs fell behind the code they describe.
Read →Documentation that lives in Markdown files and code comments becomes a navigable layer of the code graph, connected to the structural nodes it describes, searchable by meaning, and always consistent with the current codebase state.
Read →Every code intelligence tool that uploads your source to the cloud is a potential compliance violation. The source-free brain model gives you full graph intelligence without ever exposing your code, here's how.
Read →A tour of the 28 MCP tools Spiderbrain serves: 17 codebase tools (blast radius, scoring, community, insights), 9 memory tools (recall, comment, scope), and 2 blueprint tools. Which ones to reach for first.
Read →AI tools guess well. But a code graph that gives you a different answer each run is not a foundation you can build on. Here's why reproducibility is the moat, and what it makes possible.
Read →Save a file. Instantly see the exact downstream files that need re-checking. Not a list of all importers, the specific files whose behaviour could change because of what you just saved. This is the alert that changes how engineers work.
Read →Every codebase has keystones, the files where a production incident is most expensive. They're not necessarily the most-imported, and they're not obvious from the folder structure. Spikescore finds them.
Read →Vector search finds code that looks like your query. Graph traversal finds code that is structurally connected to your query. You need to know which one to reach for, and when you need both.
Read →