YBacked byY Combinator
abhigyanpatwari/GitNexus | Trendshift

The nervous system for agent context.

GitNexus indexes any codebase into a knowledge graph of every dependency, call chain, cluster, and execution flow, so AI agents never miss code.

GitNexus

The problem

Your agent sees files. Not the system.

A real codebase is tens of thousands of symbols tangled across files, packages, and repos. Agents that grep their way through it miss dependencies, break call chains, and ship blind edits.

262,322
symbols
1,030,565
relationships
1
repo (vscode)
SWE-bench Verified · 162 paired runs

Agents that read flows, not files.

GitNexus lets the model walk resolved execution flows instead of grepping file after file. On SWE-bench Verified (a single-repo benchmark that can't even exercise multi-repo graphs or impact analysis), reading process-by-process alone held the solve rate at exact parity while cutting 30% of tokens. The savings are a side effect of reading code the right way.

baseline
155.5M tokens
+ gitnexus
108.2M tokens
116/162 solved by both · −47.4M tokens · −931 API calls
Step 1: Index

Every relationship, resolved deterministically.

GitNexus parses every file with Tree-sitter and resolves imports, call chains, field types, and return types across the codebase: user.address.getCity().save() resolves at every hop. No embedding guesswork. A knowledge graph, precomputed.

4
hops resolved
0
guesses
100%
deterministic
CALLSIMPORTSEXTENDSIMPLEMENTSACCESSES
Step 2: Cluster

Clustering reveals the architecture.

Leiden community detection groups symbols into functional clusters, scored by cohesion and modularity. Your auth flow, your billing engine, your ingestion pipeline. Discovered, not documented.

auth-core
0.87 cohesion
ingestion
0.82 cohesion
billing
0.74 cohesion
MEMBER_OF · reason: leiden-algorithm
Step 3: Analyze

Know the blast radius before you edit.

Change one function and GitNexus traces every downstream caller, grouped by depth with confidence scores. detect_changes maps your git diff to affected execution flows before you commit.

depth 1
12 symbols
depth 2
21 symbols
depth 3
14 symbols
1 change → 47 affected symbols · 6 execution flows
Step 4: Scale

Impact analysis across repositories.

Group repos into one unified graph. Cross-repo edges connect your API to its consumers, so a breaking change in one service surfaces in every dependent repo before it ships.

3
repos
1,816
cross-repo edges
1
unified graph
@trpc-group · 3 repos · 1,816 cross-repo edges
The payoff

One query. The whole system.

Agents query hundreds of repos as one graph and get process-level answers: callers, execution flows, blast radius, in a few compact lines instead of pages of file dumps. More accurate context, radically fewer tokens, and an answer to what breaks before it does.

7
MCP tools
6
editors
index, reused forever
querycontextimpactdetect_changesrenamecypher
Works with Claude Code · Cursor · Antigravity · Codex · Windsurf · OpenCode
Fig. 2.0: the full instrument

Every analysis, precomputed.

See it work

The graph explorer, live on a real codebase: every cluster, call chain, and execution flow, queryable.

agent session · gitnexus mcp
demo.mp4
microsoft/vscode
trpc/server
fastapi/fastapi
calcom/cal.com
vercel/ai
mindsdb/mindsdb
Claude Code
Cursor
Codex
OpenCode
Windsurf
Antigravity
Fig. 1.1: many repos, one graph, every agent
262,322 symbols 12 relevant

Benchmarked on its weakest turf.

SWE-bench Verified is single-repo issue solving; it can't measure what GitNexus is actually for: querying hundreds of repos as one graph, blast-radius analysis, taint tracking, knowing what breaks before it ships. All it can see is how an agent reads code. Reading process-by-process instead of file-by-file was enough to hold the solve rate at exact parity while cutting 30% of tokens. The capabilities are the product; the savings came free.

SWE-bench Verified162 paired runs
Solve rateIdentical · 116/162 each
GitNexus
71.60%
Baseline
71.60%
Chunk deltas+3 · +1 · −3 · −1Net 0
[ Fig. 01 / Solve parity ]
Token consumptionWhere the savings come from
Total tokens−30.4%
GitNexus
108.2M
Baseline
155.5M
API calls−13.3%
GitNexus
6,091
Baseline
7,022
Saved 47.4M tokensSolve rate ±0.0 pts
[ Fig. 02 / Tokens ]
Read the benchmark

The internet noticed.

GitNexus is open source and one of the fastest-growing code intelligence projects on GitHub.

.
k+
GitHub stars on the open-source engine
4.9kForks
60k+Weekly npm downloads
abhigyanpatwari/GitNexus
Top AI Product
“GitNexus turns your codebase into a knowledge graph, and your AI agent will thank you.”
— on the fastest-growing code-intelligence engine on GitHub
Live

Developers are building with GitNexus

Join the Discord
Reddit
I think its a brilliant idea. I am an inventor in this space, being the principle investigator on the forerunner patent for GraphRAG. In the long run, its the only way to reliably auto-code.
DO
David Ostby
Co-founder at ViperPrompt · GraphRAG patent investigator
LinkedIn
I stumbled on GitNexus and gave it a try running locally. It is fast, looks stunning, and very informative. I am in awe how after years of AI hype, you can still be surprised by what is possible.
FB
Frank Buters
GenAI & Data Leader, Financial Services
Reddit
We're using GitNexus blast radius analysis before every code change. A voice pipeline refactor once silently broke graph memory writes: two unrelated-looking systems sharing a dependency we didn't know about. GitNexus would have caught it.
S
Stephen
Solo developer · Neura, 51+ subsystems
X (Twitter)
GitNexus exposes an MCP-driven knowledge graph directly in the browser. It's all local, nothing leaves your machine, while you still get Cypher graph queries, BM25, impact analysis, and more.
AL
André Lindenberg
Developer · @andrelindenberg
Reddit
Really impressive work. The phased ingestion pipeline is cleanly separated: structure, parse, resolve, relate, cluster, trace, embed. Leiden community detection on the relationship graph is a smart choice.
S
SithLordRising
Knowledge Infrastructure Engineer
Reddit
Game-changer! I used to grep through every file and hope I didn't miss a dynamic import. Now I can see the actual dependency graph!
S
SioSuenos64
Developer
Reddit
I think its a brilliant idea. I am an inventor in this space, being the principle investigator on the forerunner patent for GraphRAG. In the long run, its the only way to reliably auto-code.
DO
David Ostby
Co-founder at ViperPrompt · GraphRAG patent investigator
LinkedIn
I stumbled on GitNexus and gave it a try running locally. It is fast, looks stunning, and very informative. I am in awe how after years of AI hype, you can still be surprised by what is possible.
FB
Frank Buters
GenAI & Data Leader, Financial Services
Reddit
We're using GitNexus blast radius analysis before every code change. A voice pipeline refactor once silently broke graph memory writes: two unrelated-looking systems sharing a dependency we didn't know about. GitNexus would have caught it.
S
Stephen
Solo developer · Neura, 51+ subsystems
X (Twitter)
GitNexus exposes an MCP-driven knowledge graph directly in the browser. It's all local, nothing leaves your machine, while you still get Cypher graph queries, BM25, impact analysis, and more.
AL
André Lindenberg
Developer · @andrelindenberg
Reddit
Really impressive work. The phased ingestion pipeline is cleanly separated: structure, parse, resolve, relate, cluster, trace, embed. Leiden community detection on the relationship graph is a smart choice.
S
SithLordRising
Knowledge Infrastructure Engineer
Reddit
Game-changer! I used to grep through every file and hope I didn't miss a dynamic import. Now I can see the actual dependency graph!
S
SioSuenos64
Developer

GitNexus vs grep and embeddings

GitNexus
Resolved graph: exact callers, callees, imports
grep + embeddings
Text matches and similarity guesses

Deterministic, Not GuessworkImports, call chains, field types, and return types resolved with compiler-grade static analysis, so answers about structure are exact
Precomputed at Index TimeThe graph is built once and reused forever, so agents get compact answers instead of burning tokens re-exploring your codebase
100% Local & PrivateIndexing, storage, and queries run on your machine. Nothing leaves your network. Enterprise-safe by default

Frequently Asked Questions

Have more doubts? Reach out to us at founders@akonlabs.com
Want GitNexus on your team's repos?Akon Labs (YC S26) runs GitNexus as a managed SaaS or fully self-hosted, with PR review, auto-updating wikis, and multi-repo graphs included.