Spiderbrain is a critical infrastructure component for the agentic stack, specifically occupying the memory and reasoning layer. By providing a structural context graph via the Model Context Protocol (MCP), it enables agents to move beyond simple keyword searches and actually understand the architectural hierarchy of the projects they are working on. This is particularly relevant for coding agents and autonomous developers that need to manage large, complex codebases without exceeding context limits or losing track of dependencies.
The company is a champion of local-first agentic infrastructure and standardized communication through MCP. For builders, Spiderbrain offers a way to inject 'grounded' reasoning into agents, reducing hallucinations by ensuring the agent is always aware of the most load-bearing components of a system. Its deterministic nature makes it suitable for CI/CD pipelines and enterprise environments where auditable AI behavior is a prerequisite for deployment.
Spiderbrain is built on the premise that large language models suffer from context rot—a degradation in accuracy that occurs as context windows fill with irrelevant information. The company describes its product as a cognition layer that sits above the traditional context layer. Rather than relying on Retrieval-Augmented Generation (RAG) which retrieves items based on text similarity, Spiderbrain constructs a living memory graph. This graph maps the structural dependencies of a project, identifying which files are 'load-bearing' and which are peripheral. This distinction is quantified through 'signal lift,' a metric that measures how effectively the system isolates critical information from background noise.
The core of the Spiderbrain engine is its scoring system, which includes three primary metrics: webscore, spikescore, and blastVolume. Webscore rates a node's severity, indicating how much the project would suffer if that specific component failed. Spikescore measures the blast radius, tracing how a failure in one file would spread across its dependents. BlastVolume combines these into a single risk profile. By ranking files this way, Spiderbrain allows AI agents to focus their limited attention on high-leverage code. The company reports significant signal lift across diverse datasets, from software codebases where it isolated 61 'master' nodes out of over 1,600 files, to clinical trials and insurance risk sets.
Privacy and consistency are central to the Spiderbrain architecture. The tool is designed to run locally, meaning a user's code and data do not leave their machine. The process is bit-for-bit deterministic: the same input repo will always produce the same memory graph. This is a departure from many AI tools that rely on probabilistic outputs from hosted models. Spiderbrain parses real imports across multiple languages—including JavaScript, Python, Go, and SQL—as well as documentation and chat history to ensure the graph reflects the actual state of the work rather than an approximation. Decisions made by the system are append-only and hash-chained, creating an auditable 'thought log' that users can inspect.
Spiderbrain is an early and active participant in the Model Context Protocol (MCP) ecosystem. It functions as a native MCP server, which allows it to connect directly to agentic IDEs and clients like Cursor, Claude Code, Windsurf, and Zed without custom glue code. When an agent is connected to a Spiderbrain 'brain,' it inherits the structural reasoning and persistent memory of that graph. This connection is designed to be invisible to the user once configured, allowing the agent to pull from a ranked list of relevant files rather than searching blindly through a thousand grep hits. The project was developed by Perform Digital, a specialized implementation firm that used spider neuroanatomy as a biological model for externalized dependency-graphed memory.
A living context graph that gives AI agents persistent memory and structural reasoning.
Spiderbrain is hiring.