Incremental resync
On each commit or branch switch, Auryn re-parses only the modified files and the code they affect, approximately 150x faster than a full re-ingestion.
Context and memory for LLMs and agents
GraphRAG
Ground LLM answers in your knowledge graphAgentic AI
Agents grounded in how your domain connectsAgentic memory
Queryable memory that carries context across sessionsChatbots
Context-aware chatbots with long-term memoryProven patterns on connected data
Fraud detection
Trace rings of accounts, devices and transactions liveCybersecurity
Assets, findings and threats queried as one graphSupply chain
See how one supplier delay ripples downstreamReal-time recommendations
Suggestions based on how things actually relateAuryn gives your coding agents a holistic view of every repository your teams own. All code is mapped into a single graph, so any symbol, dependency, or call path is within reach. Agents work with greater accuracy, complete tasks faster, and consume fewer tokens.
SubmitOrderButton (web-portal) and NightlyReorderJob (api-gateway) call it.
466k to 355k tokens per task.
208s to 70s per task.
92% of key facts covered, vs 22% without Auryn.
$0.72 to $0.38 per task.
without Auryn with Auryn
Most organizations split their code across many repositories and many teams. Auryn maps all of it as a single graph. When a repository references code that is not yet indexed, Auryn records it and links it as soon as the repository is added.
On each commit or branch switch, Auryn re-parses only the modified files and the code they affect, approximately 150x faster than a full re-ingestion.
Full semantic search is built in and available out of the box. No separate search service is required.
Supports Claude Code, Cursor, Copilot, and other widely used coding agents immediately after installation.
Repositories may be added in any order, from any team. When code references a symbol that is not yet indexed, Auryn records the reference and resolves it once that repository is added.
Auryn performs two levels of analysis. A fast static pass runs first and is sufficient for many codebases. A deeper analysis then runs in the background for languages such as Python and C#, ensuring that critical information in weakly typed languages is captured.
See what your agent did. It reasons with the graph, and each answer references the node, file, and edge from which it was derived.
The promo code reaches billing.charge through apply_promo.
Measured head to head on the same repos, the same commits, and the same machine.
Built in. Before an agent changes shared code, it checks who calls it, what the change reaches, and how they connect.
Your engineers can open the graph your agents query and see exactly what they relied on.
Read the code, run it on your own infrastructure, and build on it. Issues and pull requests are welcome.
View on GitHubA permissive license. Use it, change it, and ship it in your own products.
The graph lives in FalkorDB, a property graph database, not a JSON file.
Parsing runs locally. The graph lives on your machine or a FalkorDB server you run.
Install is user-scoped. Nothing is written into your repositories.
Auryn includes semantic search. The difference is what comes back: the matching code plus its calls, imports, and dependencies, so the agent can act on it without reading around.
Go to definition answers one hop. A graph answers what spans the codebase: the blast radius of a change, every path between two symbols, and everything an endpoint touches down to the tables. Your agent can query it too, not just you.
No. Search finds the code. The graph shows what that code calls, imports, and depends on, including across repos.
Closure is everything a change can reach. Blast radius is who calls it and what the change can break. Traceability is the path between them, each hop cited with the file and the edge.
They can take on work that spans repositories and languages without reading the tree first. On the same graded questions, answers covered 92% of key facts instead of 22%, agent run time dropped 66%, and cost per task dropped 47%.
Claude Code, Cursor, Copilot, Codex, Gemini, Windsurf, VS Code, and Aider, plus any agent that reads AGENTS.md.
Yes, and it’s the main use case. Add repos in any order and cross-repo references connect as each one arrives. Microservices, a monolith split across repos, and shared libraries all work.
Seconds. Auryn re-parses only the files a commit touched and the code they affect. Angular, at 5,256 TypeScript files, moves to the next release in under 8 seconds.
Yes. Point Auryn at a shared FalkorDB server and every team’s agents query the same graph.
No. Install is user-scoped and the graph lives outside your repos.
Parsing runs locally. The graph lives on your machine or on a FalkorDB server you run.
In FalkorDB, a graph database, not a JSON file. It stays loaded, so caller and impact lookups come back in about 5 ms on repos the size of Django and Angular.
One command installs Auryn and plugs it into the agents your teams already use. Add repos as you go. Your agents give more accurate answers, burn fewer tokens, and finish in a third of the time, at about half the cost.