CodeMem | Local AI memory for code

Quick Demo

30-Second Demo

Run these commands and instantly see how CodeMem surfaces relevant authentication context.

Commands

codemem init
codemem start
codemem ask "how does authentication work"

Output

Relevant Files:

login.ts
authMiddleware.ts

Flow:

User → login → JWT → middleware

Why CodeMem

Stop re-pasting code and let your tools remember the project.

Private local indexing

CodeMem runs locally, stores embeddings on disk, and never sends your code to the cloud unless you choose a hosted AI provider.

AI-agnostic

Use OpenAI-compatible APIs, Anthropic, or local transformers. Switch providers without rearchitecting your workflow.

Reduce tokens, increase relevance

Only the most relevant code chunks are retrieved for each query, keeping prompts efficient and responses accurate.

Build with confidence

Power chat, ask, plan, and apply workflows from the same sidecar agent, with built-in command and patch support.

Install

Install CodeMem in minutes.

Option A — Clone and build

git clone https://github.com/mrabhishek1105/Codemem.git
cd Codemem
npm install
npm run build
npm link

Option B — Install from npm

npm install -g @mrabhishek1105/codemem

Requires Node.js 18+ and npm 8+. Works on Windows, macOS, and Linux.

Quick Start

Start using CodeMem with your project.

1. Initialize

Run this in your project root to create the local index and download the model.

codemem init

2. Start the sidecar

Launch the local server and begin querying code from any tool.

codemem start

3. Query

Search with the CLI or via HTTP using your local code memory.

codemem search "how does auth work"
codemem ask "explain the payment flow"

Workflows

Common developer workflows

1. Understand codebase

Ask CodeMem to explain a system and reveal the relevant files in your project.

codemem ask "how payment system works"

2. Implement feature

Generate a plan, review it, and apply changes with confidence.

codemem plan "add JWT auth"
codemem apply

3. Use with AI assistants

Connect Claude, ChatGPT, or other MCP clients so AI tools call CodeMem automatically for context.

AI calls CodeMem via MCP
Gets code context automatically

Commands

Powerful CLI commands for every workflow.

codemem init

codemem init [--force] [--model <name>]

Initialize the local CodeMem index for the current repository. This command creates the `.codemem` folder, extracts code chunks, and downloads the chosen embedding model so the project is searchable and ready for AI workflows.

Options

`--force`	Rebuild the index from scratch, ignoring any existing local cache.
`--model <name>`	Choose which embedding model to use for indexing the repository.

Example

codemem init --model minilm

codemem start

codemem start [--port <n>] [--mcp]

Start the local CodeMem sidecar server, which exposes the search and AI workflow endpoints. The server also optionally enables the MCP bridge for tools like Claude Desktop and Cursor.

Options

`--port <n>`	Override the default server port (8432) for the local HTTP API.
`--mcp`	Enable MCP bridge mode for compatibility with MCP-enabled clients.

Example

codemem start --port 9000 --mcp

codemem search

codemem search <query> [--top <n>] [--json]

Search the indexed repository for the most relevant code fragments matching your query. This command helps you find where logic lives without manually scanning files.

Options

`--top <n>`	Return the specified number of top results. Default is 5.
`--json`	Output the results in machine-readable JSON format.

Example

codemem search "auth middleware" --top 10

codemem ask

codemem ask <question> [--provider <name>] [--top <n>]

Ask an AI question about your repository using the stored semantic code context. CodeMem retrieves the most relevant chunks and sends them to your chosen provider for an accurate answer.

Options

`--provider <name>`	Select the AI backend: `openai`, `anthropic`, or `local`.
`--top <n>`	Choose how many indexed context chunks to include in the query.

Example

codemem ask "how does login work" --provider anthropic

codemem plan

codemem plan <task> [--out <file>]

Generate an AI-driven implementation plan for a requested task. CodeMem saves the plan so you can review it or use it later with the apply workflow.

Options

--out <file> Save the generated plan to a file instead of using the default `last-plan.json` location.

Example

codemem plan "add OAuth login" --out plan.md

codemem apply

codemem apply [] [--dry-run] [--test]

Apply a previously generated plan to your repository by producing file patches and offering a review step. This command can preview changes and optionally run tests after applying them.

Options

`--dry-run`	Show the patch diff without writing any files.
`--test`	Run tests after the patch is applied to verify the result.

Example

codemem apply plan.md --dry-run

Release Highlights

Features that make CodeMem compelling.

Local-first architecture

Keep code private and control exactly which queries reach hosted AI providers.

Semantic code retrieval

Index code once and retrieve the right context for each question, not entire files.

MCP bridge support

Use CodeMem with Claude Desktop, Cursor, and other MCP-compatible tools.

Self-hosted server

Use the local HTTP API to integrate CodeMem into your own developer tooling.

AI Integration

How CodeMem connects to AI tools.

Mode 1 — CLI

Run direct commands like `codemem ask` from your terminal to get answers based on project context.

codemem ask "what does auth do"

Mode 2 — MCP

Use Claude, Cursor, and other MCP-compatible assistants. They call CodeMem automatically and receive relevant code context.

Mode 3 — Browser AI

Generate full results from AI with output mode flags and copy code for browser-based workflows.

codemem ask "fix auth" --output=full

Output Modes

Choose how CodeMem returns results.

Context

Get an explanation plus the relevant code context for your query.

codemem ask "fix auth" --output=context

Patch

Ask for a complete file patch so you can apply changes directly or review them first.

codemem ask "fix auth" --output=patch

Full

Request a full response that includes expanded context, code, and implementation guidance.

codemem ask "fix auth" --output=full

How it works

How CodeMem works in four steps.

Index your code using AST parsing and embeddings.
Retrieve the most relevant code chunks for each query.
Build structured context from project files.
Send the context to AI for an accurate answer or patch.

API

Use the local HTTP API.

Integrate CodeMem into your tooling with a simple POST request.

curl -X POST http://localhost:8432/api/v1/query \
  -H "Content-Type: application/json" \
  -d '{"query":"auth flow"}'

Troubleshooting

Quick fixes for common issues.

Server not starting

Run the sidecar and confirm the process is listening on the expected port.

codemem start

Port issue

Use the environment override when the default port is unavailable.

CODEMEM_PORT=9000 codemem start

Index error

Reset the local database and reinitialize the repository.

rm -rf .codemem/db
codemem init

Configuration

Customize your CodeMem setup.

Use `.codemem/config.json` to configure the local project settings.

{
  "port": 8432,
  "model": "minilm",
  "provider": "openai"
}

Project structure

What CodeMem stores in your repo.

.codemem/
  db/
  meta/
  config.json

Limitations

What to expect from CodeMem.

Works best with well-structured codebases and clear project layouts.
Large repositories may take longer to index and require more local storage.
AI responses depend on the quality of the selected provider and model.

About the developer

Built by Abhishek Kumar

Founder & engineer

Abhishek builds infrastructure and AI-powered developer tools that scale with production systems. He designs solutions for CI/CD, observability, cloud infrastructure, and semantic code intelligence.

What he works on

AI & LLM integration for developer workflows
Semantic search, AST parsing, and code indexing
DevOps, monitoring, and cloud automation

Learn more about Abhishek’s work on GitHub and his portfolio at abhishekk-three.vercel.app.

Ready to deploy

Host the docs on GitHub Pages from this folder.

Use `codemem_docs` as your Pages source, and users can read the release notes, install instructions, and command reference in one place.

Open CodeMem repo Back to top