Exact files, lines and callers for your agent.

Sieve indexes a repo into a symbol graph and answers with the file and the line. It runs on your machine. No API key.

$ curl --proto '=https' --tlsv1.2 -LsSf https://github.com/iamalvisng/sieve/releases/latest/download/sieve-cli-installer.sh | sh

$ brew install iamalvisng/tap/sieve

$ cargo binstall sieve-cli

macOS and Linux.

Read the quickstart GitHub

vite · 1,583 files

$ sieve ask "how does the dev server handle hot module replacement"ask  how does the dev server handle hot module replacement · 8 hits 1  [packages/vite/] createServerModuleRunner  fn  packages/vite/src/node/ssr/runtime/serverModuleRunner.ts:143-160   function createServerModuleRunner( environment: DevEnvironment, options: ServerModuleRunnerOptions = {}, ): ModuleRunner 2  [packages/vite/] handleHMRUpdate  fn  packages/vite/src/node/server/hmr.ts:411-679   async function handleHMRUpdate( type: 'create' | 'delete' | 'update', file: string, server: ViteDevServer, ): Promise<void> 3  [packages/vite/] _createServer  fn  packages/vite/src/node/server/index.ts:513-1148...

Three commands you will use every day.

Each answer names a symbol, its kind and an exact span. Your agent opens the span and reads nothing else.

Ask in plain words. Get ranked symbols with the file and the line.

$ sieve ask "how does the dev server handle hot module replacement" --sourceask  how does the dev server handle hot module replacement · 8 hits 1  [packages/vite/] createServerModuleRunner  fn  packages/vite/src/node/ssr/runtime/serverModuleRunner.ts:143-160   function createServerModuleRunner( environment: DevEnvironment, options: ServerModuleRunnerOptions = {}, ): ModuleRunner ```export function createServerModuleRunner(  environment: DevEnvironment,  options: ServerModuleRunnerOptions = {},): ModuleRunner {  const hmr = createHMROptions(environment, options)  return new ModuleRunner(...

See who calls a symbol, to the depth you choose. Each edge says exact or inferred.

$ sieve callers handleHMRUpdate --depth 2handleHMRUpdate  fn  packages/vite/src/node/server/hmr.ts:411-6792 callers · 5 within 2 hops├─ testRestartDuringHotUpdate  fn  packages/vite/src/node/server/__tests__/hmr.spec.ts:6-40│  │  34: handleHMRUpdate(│  └─ hmr.spec.ts  file  packages/vite/src/node/server/__tests__/hmr.spec.ts:1-56└─ onHMRUpdate  fn  packages/vite/src/node/server/index.ts:908-915   │  913: await handleHMRUpdate(type, file, server)   ├─ onFileAddUnlink  fn  packages/vite/src/node/server/index.ts:917-953   └─ onFileChange  fn  packages/vite/src/node/server/index.ts:955-969[sieve] saved ≈ 20,282 tokens

Run it on a diff. See what depends on the changed lines, and whether a test reaches them.

$ sieve blastyour diff changes handleHMRUpdate · 1 symbol in 2 files · 1 of 1 reached by a test · working tree vs HEAD3 callers affected within 2 hops└─ onHMRUpdate  fn  packages/vite/src/node/server/index.ts:908-915   ├─ onFileAddUnlink  fn  packages/vite/src/node/server/index.ts:917-953   └─ onFileChange  fn  packages/vite/src/node/server/index.ts:955-9691 test suite also references this code (not listed)tests for handleHMRUpdate were not updatedask for review  翠 (80 commits, 11d ago) · btea (4 commits, 28d ago)not indexed  1 file: .gitignore

Code says what. Records say why.

sieve why links a symbol to the decision records that explain it, plus the tests that cover it. A record is an ADR file or a LEDGER.md entry with a - **Code:** line. Your existing ADR folder works as it is.

$ sieve why savings_linewhy savings_line  crates/sieve-savings/src/lib.rs:96decision  docs/adr/adr-0004-savings-header.md:1  ADR 0004: Each query result opens with one savings line          link: Code anchortest  crates/sieve-savings/src/lib.rs:236  savings_line_names_the_given_producttest  crates/sieve-savings/src/lib.rs:248  test_savings_sieve_header_has_no_tally_instruction...

Real output from the Sieve repo.

How why reads your records

A pane for Claude Code, and six small friends.

Run sieve init --mod, then /sieve-pane. The pane shows callers, impact, decision records and saved tokens next to your session.

A status line keeps a pixel mascot in view. Sieve, the kitchen sieve, is the default. Run /sieve-mascot soya if you prefer the poodle.

Set up the pane

Pick a mascot

> /sieve-mascot sieve

Sieve×

rate-limit.tssrc · read

SAVED TODAY61.7kTHIS WEEK2.3kIMPACT3

MOST CALLED SYMBOLScheckApiKeyRateLimit:313 3 callersresetWindow:40 2 callerskeyOf:12 1 callerWHO DEPENDS ON THIS◉ checkApiKeyRateLimit└─● enforceApiKeyRateLimit guard.ts:24 ├─○ POST ai/chat └─○ POST ai/job-fitIMPACT████░░░░░░░░░░ 3 callers · 2 hopsWHY IT EXISTS▍2026-09-16 rate limit per API keySAVED · LAST 7 DAYS▇▇▁▁▃▃██▅▅▁▁▂▂

It stays on your machine.

Read ADR 0001 and the privacy page.

No network call. A query never leaves your laptop.
No API key. No model pass. Your code is not sent to a provider.
No telemetry. sieve init writes only inside the repo unless you pass --global.

Numbers you can repeat.

Index 1,583 files from scratch: 1.83s
Re-index with no change: 0.80s
Median answer to a question (185 ms max): 158ms
of exact cross-file calls match the TypeScript compiler (1,232 of 1,232, name‑level): 100%

vite · 1,583 files · Apple M1 Pro · Sieve 0.1.0 · method →

Limit Sieve prints a token estimate after each query. Token savings in a real agent session are not measured yet.

Any MCP host.

sieve mcp serves the index over stdio with seven tools: find code, find all, trace calls, file API, repo map, check freshness and why.

> { "mcpServers": { "sieve": { "command": "sieve", "args": ["mcp"] } } }

The seven tools

Three lines to your first answer.

$ curl --proto '=https' --tlsv1.2 -LsSf https://github.com/iamalvisng/sieve/releases/latest/download/sieve-cli-installer.sh | sh

$ brew install iamalvisng/tap/sieve

$ cargo binstall sieve-cli

macOS and Linux.

$ sieve build .

$ sieve ask "how does login work"