Your AI keeps searching your codebase.It should just know it.

Dejavu gives Claude a memory of your codebase: semantic search, persistent session recall, and full dependency visibility across sessions.

Works with Claude Code·Tested on messy real-world repos·Free tier included

Why coding agents waste your time

It keeps searching the same repo over and over

Claude greps, globs, and re-reads files because it has no persistent understanding of your project.

It forgets yesterday's work

Architectural decisions, gotchas, and patterns disappear every session, so you keep re-explaining the same codebase.

It changes code without seeing the blast radius

Without callers, importers, and dependency context, a “small” refactor can quietly break half the app.

What changes when your AI actually knows the codebase

Without Dejavu
→ glob **/*.ts             200 results
→ grep "auth"              47 matches
→ read file 1
→ read file 2
→ read file 3
→ grep "session"           22 matches
→ read file 4
→ read file 5
→ grep "middleware"         8 matches
→ read file 6
→ read file 7
→ read file 8
  ...still looking...
40–100 tool calls. Thousands of tokens. Maybe finds it.
With Dejavu
→ dv_vision("auth session flow")

  1. src/middleware/auth.ts      (82%)
  2. src/services/session.ts     (74%)
  3. src/api/routes/login.ts     (68%)

  Ranked by semantic relevance.
One call. Real entry points. No blind wandering.

Tested on real, messy repos — not toy demos.

Built for real repos, not clean demos

Codex repo

Used to retrieve relevant files and project context without manual grep loops.

Documenso

Handled a real production codebase with cross-file relationships and messy structure.

90k LOC internal app

Tested on a large evolving codebase where session memory and safe refactors actually matter.

Your AI gets better every session

Session 1First index

Indexes the repo and learns the basic architecture.

Session 3Recalls context

Remembers what not to touch and where shared logic lives.

Session 7Safe refactor

Finds all call sites before refactoring, instead of guessing.

Session 12Team onboarding

A new engineer can load weeks of accumulated context in one recall.

Why Dejavu exists

Claude can write code. It still does not:

  • Retain project memory across sessions
  • Understand your repo semantically by default
  • Map dependencies before making changes

Dejavu adds the missing layer: persistent context for real codebases.

Two minutes to make Claude project-aware

1

Install Dejavu

npm install -g @nfinitecontext/dejavu
2

Add it to Claude Code

claude mcp add dejavu \
  -e DEJAVU_API_KEY=YOUR_KEY \
  -- dejavu-cli
3

Index your repo

> "Index this project"

dv_absorb → 847 files indexed
dv_recall → context loaded
dv_vision → semantic search ready

Works on existing repos. No code changes required.

Everything your coding agent needs to stop guessing

Understand the codebase

dv_visionSemantic search by meaning
dv_symbolsFind functions, classes, interfaces
dv_structureProject tree + symbol overview

See the blast radius

dv_callersWhat calls this function?
dv_calleesWhat does this function call?
dv_importersWhat depends on this file?
dv_depsFull dependency tree

Remember what matters

dv_recallLoad all context (start of session)
dv_imprintSave session summary (end)
dv_distillCompress old sessions
dv_etchSave architectural insights

Keep the index fresh

dv_absorbIndex an entire codebase
dv_syncRe-sync a single changed file
dv_filesList all absorbed files
dv_statusCheck memory health

Pricing

Free

For trying Dejavu on one project

$0 forever
  • 1 project
  • 200 files
  • 50 searches / day
  • All tools included
  • Session memory
  • Dependency graph

Pro

For people coding with Claude daily

$12/mo
  • Everything in Free, plus:
  • Unlimited projects
  • 5,000 files
  • Unlimited searches
  • Priority indexing

Team

Shared memory for teams working in the same codebase

$39/seat/mo
  • Everything in Pro, plus:
  • 20,000 files per seat
  • Shared project access
  • Team notes & sessions
  • Priority support

Stop paying for the same
context twice.

Your AI should not spend 20 minutes rediscovering code it already saw yesterday.