Precise context capsule
It hands your agent exactly the code each task needs — typically 4–8k tokens instead of 20–50k. Shorter prompts, lower bills, more done before you hit the limit.
What you get
Your agent stops guessing — and edits the right files the first time.
AIndex maps how your code connects and hands your agent only the code each task needs. Fewer wrong edits, fewer hallucinations, up to 80% fewer tokens — 100% on your machine.
Context pipelinerepo → graph → agent
Before / after
Same task. A fraction of the context.
Without AIndex20–50k tokens
Your agent reads whole files, guesses how they connect, and often edits the wrong place.
With AIndex4–8k tokens
Your agent gets exactly the connected code each task needs — and edits the right files the first time.
What it does
Five capabilities your agent gets the moment you install.
Search anything
Find any symbol by name, kind or meaning across the whole project in about 15ms.
Impact before you edit
Know what a change will touch before your agent edits — so it fixes the right thing without quietly breaking others. It even flags dead code and tangled spots worth cleaning up.
Cross-session memory
Insights and decisions persist between sessions, and automatically go stale when the underlying code changes.
Works with every agent
One setup command wires it into Codex, Claude Code, Cursor, Windsurf, VS Code and Zed.
100% local
Your code is indexed on your machine. Nothing is uploaded, no telemetry — only license activation touches the network.
What ships
A real local engine, not a loose prompt pack.
You get a signed binary, the MCP server with five tools, one-command agent setup, and machine-readable JSON for automation — all running locally.
Native agent wiringone setup → every client
Five tools, one server
context, search, navigate, remember and status — your agent calls them directly over MCP instead of asking you to run commands.
Real dependency graph
Knows which code matters, what a change will affect, and where dead weight and duplication hide.
One setup command
Codex, Claude Code, Cursor, Windsurf, VS Code and Zed get generated config — with rollback if you change your mind.
Private project memory
Your index, project profiles and cross-session memory live in a local store on your machine — with import/export to move between machines.
Native agent workflow
You install once. The agent uses it quietly.
AIndex should remove work, not add ceremony — so it indexes 10k files in about 8 seconds and answers a context query in about 15ms.
- Install
One command verifies the Ed25519 signature and checksum, then places the binary on PATH.
- Set up the repo
AIndex detects the project, builds the dependency graph and writes agent config — re-indexing incrementally in about 50ms after edits.
- Ask your agent normally
The agent pulls the exact connected context through MCP tools instead of relying on pasted files.
- Verify and recover
status, JSON outputs and clean uninstall paths make the index health visible and every step reversible.
Why it matters
Less guessing. Fewer tokens. Code that holds together.
Agents that read the graph, not just the file
Your agent works from the code that actually matters for the task — the real, connected pieces — instead of whatever file happened to be open.
Up to 80% fewer tokens per task
Tiered context means smaller prompts, lower bills, and far more work done before you hit the context limit.
Private by default
No hosted index, no source upload. The engine runs entirely on your machine; only license activation ever talks to the network, and Free is fully offline.
Stop letting your agent guess.
Download AIndex, point it at a repo, and your agent gets the exact context in one command. Free means no account, no license and no card — your code never leaves your machine.