A napkin sketch of why your
AI bill is fat.
Most agents read like they're being paid by the token. Turns out, they are.
Without RAG
- 01agent greps randomly
- 02reads 12 irrelevant files
- 03burns 8k tokens
- 04still asks clarifying q's
With Ragtoolina
- 01semantic retrieval
- 02reads 2 relevant chunks
- 032.8k tokens — done
- 04one-shot answer
Three steps. That's it.
No infra changes. No new workflow. ~3 minutes.
Download & launch
Install the macOS app. Runs locally — your code never leaves your machine.
Add your projects
Point at any folder. Ragtoolina indexes symbols + semantic meaning in seconds.
Connect & code
Any MCP-compatible tool. Add one line of config. Done.
// .cursor/mcp.json { "mcpServers": { "ragtoolina": { "command": "ragtoolina", "args": ["mcp"] } } }
Measured, not marketed.
Real benchmark on Cal.com (40k⭐) with Claude Sonnet 4. No cherry-picking.
We publish the bad row too.
Real queries run on Cal.com. Q1 shows overhead — we're not hiding it.
| Query | Without | With Ragtoolina | Savings |
|---|---|---|---|
| Find a specific exported functionsimple lookup — slight overhead | 1,240 | 1,510 | +22% overhead |
| Explain the authentication flow | 8,520 | 2,810 | 67% saved |
| Refactor a complex component | 12,340 | 4,150 | 66% saved |
| Debug an API endpoint | 9,780 | 3,240 | 67% saved |
| Add a new feature across files | 7,170 | 2,520 | 65% saved |
| Average across all queries | 7,810 | 2,846 | 63% saved |
Drag the dial. Do the math.
Based on average token use per request, 22 workdays / month.
yearly: $3,996
Works with your tools.
Any MCP-compatible AI tool. Plug in once, save everywhere.
Cursor
AI-native IDEClaude Code
CLI agentWindsurf
AI code editorVS Code
via Cline / RooClaude Desktop
desktop appGitLab
git platformsupports anything that speaks the Model Context Protocol
Deploy on your infra.
For teams whose code has opinions about leaving the building.
Your infrastructure
On-prem or private cloud. No data leaves your network.
Full control
Manage access, storage, and updates on your terms.
Compliance-ready
SOC 2, HIPAA, internal security requirements — handled.
Simple. Four tiers.
Start free. Upgrade when it hurts not to.
Free
For personal projects.
- ✓1 project
- ✓10,000 chunks
- ✓Semantic search
- ✓MCP server
- ✓Community support
Pro
For devs who ship daily.
- ✓5 projects
- ✓100,000 chunks
- ✓Advanced search
- ✓Git history indexing
- ✓Priority support
- ✓API access
Team
Shared context for a squad.
- ✓10 projects
- ✓1M chunks
- ✓Team workspace
- ✓Shared indexes
- ✓Admin controls
Enterprise
Big fish. Big compliance.
- ✓Unlimited everything
- ✓Self-hosted deploy
- ✓SSO / SAML
- ✓Dedicated support
- ✓Custom SLA
Probably the ones you're thinking.
Can't find yours? Email us — one of us will reply.
Does my code leave my machine?
Never. Embeddings and indexing happen locally. We literally can't read your code.
Which languages are supported?
TypeScript, JavaScript, Python, Go, Rust, Swift, Kotlin, Java, C++, Ruby, PHP. More monthly.
Does it slow down my IDE?
No. Ragtoolina runs as a separate local service. Your IDE just asks it questions.
How is this different from Cursor's @codebase?
Tool-agnostic. The same index powers Cursor, Claude Code, Windsurf — no re-embedding per tool.
Start saving tokens today.
Download the app. Add your project. That's it.
takes 3 minutes · pinky promise