What is Spiderbrain?
Spiderbrain, sometimes searched as Spider Brain, is the cognition layer for AI coding: a living map of your codebase that gives AI agents persistent memory and tells them exactly what a change will break.
Most AI coding tools forget. The moment a chat window closes, the context is gone, and the next session starts from zero. Spiderbrain fixes that. It ingests every file, dependency, decision and change into one always-current graph, then serves that graph to your AI agents so they reason over the real structure of your project instead of re-reading text every time.
The problem it solves
A large codebase is mostly invisible relationships. Which files are load-bearing? What breaks if you change this one function? Which decisions are going stale? Humans lose this map over time, and language models never had it. Spiderbrain makes the structure explicit and keeps it current, so neither you nor your agent is flying blind.
How Spiderbrain works
- It builds a context graph. Every file and dependency becomes a node and an edge in one deterministic graph, parsed natively on your machine with no Node or extra runtime to install.
- It scores what matters. The graph is scored on our EU servers, where each node gets a severity score (how much its failure would hurt) and a blast radius (how much breaks along with it). Your source never leaves your device.
- It finds the keystones. The few hubs everything leans on are surfaced as masters, the natural firebreaks in a cascade.
- It serves agents over MCP. The graph drops into Claude, Cursor, Continue and any Model Context Protocol client, so agents inherit memory and reasoning that stays current.
Agentic chat that understands your codebase
Spiderbrain includes an in-app agentic chat. Unlike a generic coding assistant, it is grounded in your brain, so it knows the shape of your project and what your change will break before it makes it.
It knows your codebase
The chat retrieves the files that matter, ranked by importance (webscore), so the AI answers with the right context instead of guessing.
It warns before it breaks things
Before any edit, it shows the blast radius, how many files depend on what you are about to change, and waits for your approval.
Your models, your keys
Multi-provider: Anthropic, OpenAI, Google, or local models through Ollama and LM Studio. Bring your own key, or keep chat inference private with local models.
Ask, or Agent
Switch between read-only questions and an agent that proposes and applies scoped edits, each behind a diff with one-step undo.
Blueprint: plan it as a drawing
Every Spiderbrain project carries a Blueprint, an edgeless canvas where you sketch the architecture as blocks and wires instead of describing it in prose. Draw the system you want to exist, or let Spiderbrain draw the architecture of a freshly mapped project for you and refine it from there.
- From drawing to instruction. Share the Blueprint to your AI over MCP (a Pro feature), and the agent reads the same boxes and wires you drew, so "build this" finally has a picture attached.
- From code to drawing. A built brain can auto-draw its own architecture, so you start from the truth of the codebase, never a blank page.
- You stay the editor. Updates flow when you pull them. Nothing redraws or overwrites your plan behind your back.
Memory tree: what your AI remembers
The Memory tree is your project's decision memory: decisions, constraints, patterns and preferences, anchored to the parts of the graph they belong to and carried across sessions. It fills in as your AI works, so the reasoning that shaped the project stops evaporating when a chat window closes.
- Readable, not a black box. Open the Memory tree surface on any project and see exactly what is remembered, grouped by type and status.
- Steerable. Comment on any memory to ask your AI to change it. Requests queue until the agent applies them, and nothing changes silently.
- Hosted and private. Memory lives encrypted on our EU servers (a Pro feature) and is served to your agents over MCP, so every session starts already knowing the project.
How Spiderbrain helps you
Agents that remember
Persistent memory across every session, not just the current chat window.
Know what breaks
Severity and blast-radius scores tell you the real impact of a change before you ship it.
Find what matters
Importance scoring surfaces the handful of files that carry the whole project.
Easy to keep current
Rebuild in one click, with drift detection flagging what changed. No manual re-indexing, no stale embeddings. Pioneer founding members get live watch: the map updates as files change.
Source stays private
Parsing is native and on-device; only a source-free graph is scored. Your source code never leaves your device.
Plugs into your tools
MCP-native: Claude, Claude Code, Cursor, Continue, Zed, Windsurf and more, with no glue code.
Who it is for
Developers and teams who use AI agents and want them to actually understand the codebase, not just autocomplete it. It suits a solo engineer keeping one project coherent, and it scales to a team that needs one shared, deterministic brain. It also works on any corpus, not only code: docs, records and decisions all fit the same graph.