Guides · measured July 2026
How to make Claude Code faster.
Most of the advice is true (smaller context, the right model, tighter prompts) and all of it helps at the margin. But on a real 4.47-million-file machine we measured where an agent's session actually goes, and one thing dwarfs the rest: most of the wall-clock and about 58% of the tokens are spent finding files, not thinking about them. Fix that and Claude Code gets faster in a way no bigger subscription can buy.
TL;DR · Key takeaways
The usual advice (leaner context, the right model, tighter prompts) only trims overhead. The real drain is file search: most of the wall-clock and about 58% of session tokens go to finding files, not thinking about them.
Claude Code's fastest search tool is ripgrep, and a grep has no index: it re-walks the disk on every query. One heavy content query takes 93.8 s; the same query on a prebuilt index returns in 16 ms (58,625× faster).
Hand the agent an always-fresh index and one real file-find task collapses from 6m 57s across 71 tool calls to 16 ms in a single call, about 7,200,000× faster, with the ~58% token tax dropping to ~0%.
Filename lookups land at 85 µs (139 µs median on a 4.47M-file drive), roughly 412,000× faster than VS Code and ~480,000× faster than Windows Search (67 s). A saved file is searchable in ~1 ms; at rest the engine sits around 44 MB.
Everything local is free forever, no card. Hosting starts at $5.99/mo.
Six levers, roughly in order of impact
Five save seconds. The sixth changes the session.
None of these is a trick. They are the honest set of things that shorten a Claude Code session, ordered by how much they actually move the clock. The first five are hygiene. The sixth is infrastructure, and it is the one worth measuring.
Trim the standing context.
Every line of CLAUDE.md and every pre-pasted file is re-read on each turn. Keep it to what the model actually needs; a lean prompt plans faster than a fat one.
Match the model and effort to the task.
Maximum thinking on a one-line rename is latency you paid for and didn't need. Save the deep reasoning for real design work; let routine edits run light.
Say where, not just what.
“the auth guard in packages/api” reaches the answer in fewer turns than “find the auth guard.” A scoped path is the cheapest hint you can give.
Let it work in parallel.
Independent reads and edits can fire in one batch. Sequential tool calls that didn't need to be sequential are dead wall-clock.
Checkpoint long sessions.
Compact or restart before the context fills with stale directory listings the model will only re-read. A fresh window is often faster than a bloated one.
Give it a real file index.
This is the big one. The five above trim overhead; this one removes it: the agent stops shelling out to grep the disk and asks a prebuilt index instead. The rest of this page measures it.
One real task, measured on video
We watched Claude Code find one file.
One file-find request on the benchmark machine, resolved to a single content query. With no index, the agent reached for its shell tools and ground through it. With the same query answered by a prebuilt index, the work collapsed. Same model, same machine, same request. Here is what changed.
Tokens spent grepping are tokens not spent reasoning. Even the simplest version of this (one file, searched by name) took the agent 193.5 s of blind walking; the index returns it in 85 µs. That single lookup on its own is 2,276,000× faster, and the full task collapses by more than seven-million-fold when the answer is one indexed lookup instead of seventy-one filesystem walks. The model didn't get smarter. The floor it stands on got faster.
Anchor 1 · ripgrep
Claude Code's fastest search tool is ripgrep. That is the ceiling.
ripgrep is genuinely excellent software, one of the fastest greps ever written. But a grep has no index; it re-walks the directory tree and re-reads the bytes on every single query. When Claude Code needs to search content, it shells out to ripgrep, grep, find or Glob, so the agent's floor is ripgrep's floor. On the benchmark machine, one heavy content query took ripgrep 93.8 s; the same query against a prebuilt index returned in 16 ms. Across the five-query set, 58,625× faster on average (5.6 ms mean).
This is not ripgrep losing at its job: it is a scanner doing a scanner's work. ripgrep is the right tool when you have no index and one folder open. It is the wrong tool to run hundreds of times an hour across a whole machine, which is exactly what an agent does. The fix isn't a faster scanner. It is not scanning.
Anchor 2 · Claude Max
A bigger plan buys a smarter model. Not a faster floor.
Claude Max raises your usage ceiling and keeps the strongest model in the loop. That makes the thinking better, and it is often worth every dollar. But the retrieval layer underneath is identical on Free, Pro and Max: the same find, grep, ripgrep and Glob. Upgrading the plan does not shrink the ~58% of the session that drains into file hunting: you are paying premium rates for the model to read directory listings it forgets one turn later.
The scaffold around the model (how it retrieves, what it keeps in context, how its tools are wired) moves wall-clock more than the model tier does. Keep Max for the reasoning; fix the floor for the speed. They are different layers, and the index is the neutral one every plan plugs into.
What Max changes
Model quality in the loop
Higher usage limits
Deeper reasoning per turn
What it doesn't
How files are found
The grep-the-disk floor
The ~58% search tax
The sixth lever, set up
Hand the agent a prebuilt index. It stops grepping.
Interlinked keeps an always-fresh index of every file on the machine (names and contents, every repo, the git-ignored files a cwd grep never sees, documents included) and exposes it to any agent over MCP. One signed install auto-configures 19 AI clients, Claude Code among them, with no JSON editing. Filename lookups land at 85 µs (139 µs median across 20 queries; median, not best case); content in ~7 to 9 ms. A save is searchable in about a millisecond, under 30 ms worst case, so the index never lies about a file you just wrote. At rest the whole engine sits around 44 MB, and the index on disk stays under 1% of the drive.
The same index carries a knowledge vault that rides along on the searches the agent already runs, with no model in the loop. On a fresh machine it found 12 of 12 repos and replayed 8,455 git events in 1.7 s into 26,958 relationship pairs, no failures. Notes any agent leaves on a file are fed back to the next agent automatically. Speed finds the file; the vault knows why it matters.
ONE INSTALL · 19 AI CLIENTS · CLAUDE CODE, CURSOR, CODEX, COPILOT, WINDSURF, ZED AND THE REST.
Frequently asked
Common questions. Straight answers.
Does a bigger plan like Claude Max make Claude Code faster?
It buys a smarter model and higher limits, not a faster floor. The retrieval layer (find, grep, ripgrep, Glob) is identical on Free, Pro and Max, so a bigger plan doesn't shrink the ~58% of the session spent hunting for files. Keep Max for the reasoning; fix the floor for the speed.
Isn't ripgrep already fast?
It's one of the fastest greps ever written, but a grep has no index, so it re-walks the directory tree and re-reads the bytes on every query. On the benchmark machine one heavy content query took ripgrep 93.8 s; the same query against a prebuilt index returned in 16 ms, 58,625× faster across the query set. The fix isn't a faster scanner, it's not scanning.
How much faster is an indexed lookup, really?
A single file-find task that took 6m 57s across 71 tool calls collapses to 16 ms in one call (about 7,200,000× faster) and the ~58% of session tokens spent on search drops to roughly 0%. Filename lookups land at 85 µs (139 µs median on a 4.47M-file drive), about 412,000× faster than VS Code and ~480,000× faster than Windows Search (67 s).
Will the index be stale right after I save a file?
No. A save is searchable in about a millisecond, so the index never lies about a file you just wrote. It covers the whole machine: every repo, including the git-ignored files a cwd grep never sees, and document text too.
What does it cost, and is it heavy to run?
Everything (including agent access over MCP) is free forever, no card. Hosting starts at $5.99/mo. At rest the whole engine sits around 44 MB, less than a browser tab.
Do this today
Make Claude Code faster in one install.
Install once. One signed installer wires up 19 AI clients: Claude Code, Cursor, Codex, Copilot, Windsurf, Zed and the rest. No config, no keys.
Let it index. Filename search works immediately; content fills in behind it. At rest the whole engine sits around 44 MB, less than a browser tab.
Point it at a path when you can. Whole-machine when you don't know where something lives; scoped to a folder when you do. Scoped queries return in microseconds.
Stop pre-pasting file trees. Let the agent ask the index instead of grepping the disk. That is the move that takes the ~58% search tax down to ~0%.
Everything local is free forever, on all your devices, no card. Hosting starts at $5.99/mo. Download for Windows.