Your AI agent is smart enough.
It just doesn't know your codebase.
vexp gives your coding agent the domain knowledge it's missing. Better context in, better code out.
One-time email with install links. No spam.
20 lookups/day on free · 30 seconds to set up · No credit card
Sound familiar?
You ask Claude to fix a 3-line bug. It rewrites the entire module instead.
The agent ignores your existing patterns and reaches for an architecture you don't use anywhere in the codebase.
By message 15, the agent contradicts what it said at message 3 — and "fixes" code it already touched.
These aren't model problems. The model is brilliant.
These are context problems. The agent doesn't know which code matters, so it guesses.
Proof
Same model. Same prompt. Different context.
Task: "Fix the JWT expiry check in login"
Files in context
Files in context
Different result.
Token gauges based on internal benchmarks across 5 real codebases. File counts and test pass rates reflect typical session patterns reported by users.
How it works
Three steps. No magic.
Index your project.
AST graph + change coupling + LSP edges, built in under 15s for a 5,000-file repo. All local. Zero network calls.
Match task intent to code.
The pipeline ranks pivot files by relevance, returns skeletons (70–90% token reduction) for the supporting cast, ignores noise.
Surgical context, better output.
Your agent reads 3–5 ranked files instead of 47 random ones. Same model. Same prompt. Better code.
Works with the agents you already use:
Benchmark
On SWE-bench Verified — 100-task stratified subset, fixed seed 42, 12 repos — Claude Code + vexp resolves
of bugs on the first attempt.
7–10 issues only vexp resolves across the four agents tested.
$0.67/task — 22% cheaper than the next best ($0.86). Confidence interval ±8.7pp at 95%. The 73% vs 72% gap isn't statistically significant — we publish the full data because senior engineers deserve methodology, not marketing.
Full methodology, per-repo breakdown, raw dataThe differentiator
Your agent remembers what it learned yesterday.
Observations are linked to code symbols, not chat turns. When you refactor a class, vexp invalidates the stale memories automatically. When you come back tomorrow, your agent picks up with fresh context — not stale advice from 3 commits ago.
The only context engine where memory expires when your code does.
No competitor does this. Augment indexes commit history; vexp anchors memory to live code structure.
Adoption
9,000+ developers shipping with vexp.
Used daily across 12 AI coding agents.
Pricing
Flat pricing. No credits. No surprises.
Free
For personal projects. No account.
- 20 AI lookups per day
- Up to 2,000 nodes indexed
- 1 repo workspace
- 7 MCP tools (4 core + 3 memory)
- VS Code extension or CLI
Pro
For senior devs shipping daily.
- All 11 MCP tools (pipeline, impact, flow, skeleton, memory)
- 50,000 nodes indexed
- Up to 3 repos per workspace
- Intent detection + CodeLens
- Change coupling + cross-repo edges
- Email support
Team
Shared context for your engineering team. Min 3 seats.
- Unlimited nodes & repos
- Shared workspace + git index
- Priority paths & team analytics
- Priority support
- All Pro features included
Need a custom solution? Talk to us →
Install in 30 seconds.
Free tier. No credit card.
One-time email with install links. No spam.
Cancel anytime. Works with the agents you already use.