SpiderBrain (spiderbrain.ai) is the coding intelligence layer for AI agents: a deterministic, source-free file graph that maps how a codebase actually connects, scores every file by blast radius and severity, surfaces the keystones and blind spots that matter, and gives AI agents persistent project memory over MCP. In short, it tells an AI exactly what breaks before it ships.
No reviews yetBe the first to leave a review for SpiderBrain.ai - Built to earn its keep.
Maker
📌
Introducing Spiderbrain V3
Massive AI Hallucination and Cost Reduction (Tokens) for Claude, Cursor and other AI coding agents.
Key problems with AI coding assistants today:
1. Memory issues
They forget something about your project every session. Agent memory gets compressed repeatedly and information is lost during every transition.
2. As projects grow larger, agents hallucinate and invent file names, functions, imports, and dependencies that do not exist. Massive amounts of tokens are then spent fixing errors introduced by the agents themselves.
3. AI tools keep burning money rediscovering what they already knew last week. Sessions do not persist, memory disappears, and developers must constantly re-explain context. Testing, debugging, and bug fixing consume huge amounts of tokens.
And these are only a few of the problems.
Spiderbrain reduces:
Over 80% hallucination | 95% cross-session rediscovery costs | 77% fewer tool calls | Up to 70% faster workflows | & more
Steps to install and test:
1. Copy the GitHub URL : https://lnkd.in/gAfHfGw9
2. Paste the URL into your AI coding agent and point it to your project folder with text: "Use Spiderbrain on my project"
That's it, it'll quickly create a folder it uses to assign scores and weights to your files without effecting your project.
What it does:
Spiderbrain is built around two core concepts:
1. Inspired by how a spider's brain works, it creates a parallel intelligence layer alongside your project without modifying the project itself.
2. It generates multiple contextual variables and builds a K-nearest-neighbour cluster model using an 8-leg topology. Each node carries a webscore value that determines information traversal and routing.
Key Benefits:
1. #Reducesredundancy across AI sessions, teammates, and tools by creating one canonical project brain instead of repeated rediscovery.
2. #Reduceshallucinations by giving agents a structural ground-truth reference before generating file names, APIs, dependencies, or versions.
3. Improves #routingefficiency by providing orchestrators with a project-level amplitude signal based on mass, recency, and master phase.
4. Improves #inferenceutilisation by collapsing the rediscovery phase that consumes most session tokens.
5. Lowers operational costs through compounded efficiency gains across development workflows.
More info: https://lnkd.in/gstAwjCJ (#SHA hash available on #blockchain)
Honest disclaimers:
1. Recently releases and not tested on enterprise scale projects yet, initial test displayed the bigger the dataset the faster the algorithm works.
2. I created it for #claudecode over time 6 others are also planned including #gemini #cursor #openai & #mistral
3. The git repo is source available on #github, #challenges are in place. Source - available under BUSL 1.1, free for personal use.
Report
Maker
Don't forget to use the MCP for platforms like Opencode, Kiro, Gemini, OpenAi, Grok & others!
Can you customize how it defines clusters, or is it inferred automatically from the codebase?
Report
Maker
@naimz Currently there are three dimensions on how it works, master-client topology, rhytm (the frequency) between the files and something called webscore which is the key used to define clusters.
Webscore can be thought of as the impact of a file breaking/or being corrupted it will have on it's cluster's master.
yes, in v4 (still under improvement) we have weight definitions which a user can tweak and two others params they can use to define clusters
Don't forget to use the MCP for platforms like Opencode, Kiro, Gemini, OpenAi, Grok & others!
Mailwarm
Can you customize how it defines clusters, or is it inferred automatically from the codebase?
@naimz Currently there are three dimensions on how it works, master-client topology, rhytm (the frequency) between the files and something called webscore which is the key used to define clusters.
Webscore can be thought of as the impact of a file breaking/or being corrupted it will have on it's cluster's master.
yes, in v4 (still under improvement) we have weight definitions which a user can tweak and two others params they can use to define clusters