CodeIn

Open-source AI coding IDE — local-first, multilingual, multi-agent

Quick Start · Features · Languages · Agents · Architecture · Contributing

What is CodeIn?

CodeIn is an open-source AI coding IDE built as an Electron desktop application, VS Code extension, JetBrains plugin, and standalone web GUI — all sharing a single local agent runtime.

At its core: a Node.js HTTP server on localhost:43120 orchestrating 60 LLM provider integrations, a 13-agent autonomous swarm, Docker-sandboxed code execution, and voice-enabled prompting in 22 Indian languages. Everything runs on your machine. No cloud account required. No telemetry. Apache 2.0.

The project is at v1.0.3-beta — the architecture is complete and the core feature set works end-to-end, with ongoing polish and stabilization before v1.0 stable.

Why CodeIn?

The problem	What CodeIn does instead
Most AI coding tools require a paid subscription	Apache 2.0 — free to use, fork, and self-host forever
Cloud-only tools send your code to remote servers	Local-first: llama.cpp runs inference fully offline, on your hardware
Provider lock-in (Copilot = OpenAI, Cursor = Anthropic)	60 provider integrations — swap with your own API key, or go fully local
English-only prompting	22 Indian languages + code-mixed input (Hinglish, Benglish, Tanglish) + voice
Single-agent, single-turn chat	13-agent autonomous swarm with planning, execution, testing, review, and retry
No visibility into agent actions	Fail-closed permissions, audit logging, explicit approval gates for destructive ops

Core Capabilities

AI Engine

Feature	Description	Status
60+ LLM Providers	OpenAI, Anthropic, Gemini, Groq, Mistral, Deepseek, Ollama, Azure, Bedrock, Together, Fireworks, OpenRouter, and 48 more. Per-request cost tracking, health monitoring, failover chains.	✅
Local Inference (llama.cpp)	Fully offline inference. Qwen2.5 Coder 1.5B ships as default. No cloud dependency required to run the IDE.	✅
Multi-Agent Swarm	13 specialist agents with 4 orchestration topologies. Circuit breakers, 3-tier blackboard memory, SQLite persistence, audit logging.	✅
MCP Tool Protocol	Connect any MCP server (stdio or HTTP). Tool discovery, argument validation, JSON-RPC 2.0, audit trail.	✅
Autonomous Planner	Plan → Execute → Test → Diagnose → Revise → Retry loop with configurable retry limits and cost caps.	✅
Code Knowledge Graph	GitNexus integration for AST-backed symbol resolution, call graph tracing, and hybrid BM25+semantic search.	🔄 In progress
Vibe Coding	Upload a screenshot or describe UI intent — CodeIn generates a spec and writes production code autonomously.	🔄 In progress

Development Tools

Feature	Description	Status
Repo Intelligence	AST symbol finding, change impact analysis, semantic search, safe multi-file refactoring.	✅
Git Automation	Branching, commits, diffs, staging, PR creation — agent-orchestrated with human approval gates.	✅
Docker Sandbox	Secure agent code execution: dropped capabilities, read-only root, resource limits, network isolation.	✅
Web Research	Agents search the web, fetch documentation, cache responses, and synthesize results autonomously.	✅
Background Agents	Run agents on independent tasks in the background; monitor progress without blocking the main chat.	✅
GPU on Demand	RunPod integration: browse GPUs, spin up pods (A100, H100, RTX 4090), submit serverless inference jobs with budget caps.	✅
Computer Use	Goal-based screen automation with skill templates and audit trails. Vision pipeline for UI interaction.	🔄 In progress
Terminal Manager	Agent-controlled terminal sessions with timeout protection and cross-platform shell support.	✅

Editor and UI

Feature	Description	Status
Monaco Editor	Full VS Code editor experience: syntax highlighting, IntelliSense, minimap, multi-cursor.	✅
TipTap Rich Input	Rich text input with @mentions, slash commands, code blocks, and file attachments.	✅
xterm.js Terminal	Integrated terminal emulator with PTY support.	✅
Voice Input + TTS	Browser speech-to-text for dictation in 22 Indian languages. Backend TTS via gTTS / Azure with browser synthesis fallback.	✅
Theme Support	Dark and light themes with 50+ CSS custom properties. Follows OS preference.	✅
Plugin System	Hook-based plugin architecture with timeout protection, priority ordering, and lifecycle management.	✅

Multi-Agent Swarm

CodeIn ships a production-ready multi-agent system. Agents run as isolated tasks, communicate through a shared blackboard, and are coordinated by a swarm manager that enforces permission gates and budget limits.

13 specialist agent types:

Agent	Role
`PlannerAgent`	Task decomposition, dependency graphing, orchestration planning
`CoderAgent`	Code generation and implementation
`ReviewerAgent`	Code review and quality analysis
`TesterAgent`	Test case generation and execution
`RefactorerAgent`	Code refactoring and technical debt reduction
`DebuggerAgent`	Bug diagnosis, root cause analysis, and patching
`SecurityAgent`	OWASP-aligned vulnerability scanning
`DocsAgent`	Documentation generation from source
`ArchitectAgent`	Architecture and system design guidance
`DevOpsAgent`	CI/CD, Docker, and deployment orchestration
`BrowserAgent`	Autonomous web research and documentation retrieval
`I18nAgent`	Language detection, code-mixed input normalization, translation
`VibeBuilderAgent`	UI generation from images and natural language descriptions

4 orchestration topologies:

Pipeline         Fan-Out          Round-Robin      Hierarchical
────────         ───────          ───────────      ────────────
A → B → C        A → B            A → B → A        Orchestrator
                 A → C            A → C → A          ├── Team A
                 A → D                               └── Team B

Memory architecture: Working memory (per-task) → Blackboard (cross-agent) → SQLite persistence (~/.codein/swarm/codein.db)

Language Support

Most AI coding tools only understand English. CodeIn accepts prompts in 22 Indian languages, detects the script automatically, and translates to structured English for AI execution — so you can write the way you think.

Tier 1 — Full Support (11 languages)

UI translation · Voice input · Script detection · Text-to-speech

Language	Code	Script
Hindi	`hi`	Devanagari
Bengali	`bn`	Bengali
Telugu	`te`	Telugu
Marathi	`mr`	Devanagari
Tamil	`ta`	Tamil
Gujarati	`gu`	Gujarati
Kannada	`kn`	Kannada
Malayalam	`ml`	Malayalam
Punjabi	`pa`	Gurmukhi
Odia	`or`	Odia
Assamese	`as`	Assamese

Tier 2 — Core Support (11 languages)

UI translation · Voice input (script detection expanding)

Language	Code	Script	Notes
Urdu	`ur`	Nastaliq	Script detection not yet available
Sindhi	`sd`	Arabic	Script detection not yet available
Konkani	`kok`	Devanagari	Shares Devanagari detection
Manipuri	`mni`	Meitei	Script detection not yet available
Dogri	`doi`	Devanagari	Shares Devanagari detection
Bodo	`brx`	Devanagari	Shares Devanagari detection
Santali	`sat`	Ol Chiki	Script detection not yet available
Maithili	`mai`	Devanagari	Shares Devanagari detection
Nepali	`ne`	Devanagari	Shares Devanagari detection
Sanskrit	`sa`	Devanagari	Shares Devanagari detection
Kashmiri	`ks`	Nastaliq	Script detection not yet available

Code-Mixed Input

Write how you think. CodeIn handles natural code-mixed patterns:

What you type	What CodeIn understands
"login page bana do with Google auth"	Create a login page with Google OAuth integration
"is repo ka backend improve karo"	Improve the backend of this repository
"testing likh do is function ke liye"	Write tests for this function

LLM Providers

60 provider integrations ship in core/llm/llms/. Swap providers at runtime without restarting the agent.

View all 60 providers

Major cloud providers OpenAI · Anthropic · Google Gemini · Azure OpenAI · AWS Bedrock · AWS SageMaker · Google Vertex AI · Mistral · Groq

Secondary cloud providers Together · Fireworks · OpenRouter · Deepseek · Cohere · Cerebras · Nvidia NIM · IBM WatsonX · SambaNova · Replicate · HuggingFace Inference API

Local / self-hosted Ollama · llama.cpp · LlamaFile · LlamaStack · LM Studio · Text Gen Web UI · vLLM · Docker (containerized)

Embedding providers Voyage · HuggingFace TEI · HuggingFace TGI · TransformersJS (WASM)

Regional and specialized Moonshot · Novita · SiliconFlow · OVHcloud · Scaleway · Cloudflare · Inception · Kindo · AskSage · Nous · Venice · Mimo · NCompass · FunctionNetwork · Nebius · Lemonade · Msty · CometAPI · TARS · Flowise · Relace

Custom CustomLLM — bring any OpenAI-compatible endpoint

Skills and Automation

CodeIn ships two categories of built-in skills that extend agent capabilities beyond raw LLM calls.

Agent Runtime Skills (12)

Registered in the plugin system; available to all agents. Sandboxed with a 30-second execution timeout.

Skill	Description
`generate-commit-message`	LLM-enhanced conventional commit messages from staged diffs
`generate-pr-description`	PR summaries from diff content and commit history
`analyze-code`	Code analysis with actionable improvement suggestions
`lint-check`	Code linting analysis and style enforcement
`detect-secrets`	Scan for hardcoded secrets, API keys, and credentials
`estimate-complexity`	Cyclomatic complexity estimation for functions and modules
`generate-docs`	Documentation generation from source code
`generate-test-file`	Test file scaffolding with framework-appropriate structure
`search-codebase`	Semantic code search across the workspace
`explain-code`	Code explanation for onboarding and knowledge transfer
`suggest-refactor`	Refactoring suggestions with pattern-based recommendations
`review-security`	Security vulnerability scanning (OWASP-aligned)

Claude Code IDE Skills (11)

Located in .claude/skills/ — domain-specific guidance when using Claude Code as the AI backend.

Skill	Description
`code-reviewer`	Comprehensive code review for React / TypeScript / Node.js
`debug-helper`	Systematic debugging from reproduction to verified fix
`refactor-assistant`	Safe, incremental refactoring with test-first workflow
`security-auditor`	OWASP-aligned audit for Node.js / React / Electron
`api-designer`	REST API design covering resources, errors, pagination, versioning
`test-architect`	Testing strategy and patterns for Vitest / Jest + RTL
`performance-optimizer`	React, Node.js, and Electron performance optimization
`git-workflow`	Branching, conventional commits, PR best practices
`react-component`	React component design patterns with TypeScript and accessibility
`mcp-builder`	Build MCP servers with transports, tools, resources, and security
`docs-style`	Style guidelines for writing and reviewing documentation

Architecture

System Overview

graph TB
    subgraph Clients["Client Layer"]
        E[Electron Desktop App]
        V[VS Code Extension]
        J[JetBrains Plugin]
        W[Web GUI — React 18]
    end

    subgraph Runtime["Agent Runtime — localhost:43120"]
        direction TB
        API[27 HTTP Route Modules<br/>JWT Auth · Rate Limiting · Input Validation]

        subgraph MAS["Multi-Agent Swarm"]
            SM[Swarm Manager<br/>4 Topologies]
            AG[13 Specialist Agents]
            MEM[3-Tier Memory<br/>Working · Blackboard · SQLite]
            SM --> AG
            AG --> MEM
        end

        subgraph Compute["Compute & Inference"]
            LC[llama.cpp<br/>Local Inference]
            HUB[AI Hub<br/>60 Cloud Providers]
            GPU[RunPod GPU<br/>A100 · H100 · RTX 4090]
        end

        subgraph Tools["Development Tools"]
            SBX[Docker Sandbox<br/>Secure Execution]
            GIT[Git Automation]
            RI[Repo Intelligence<br/>AST · Semantic Search]
            MCP[MCP Client<br/>Tool Protocol]
            WEB[Web Research]
        end

        subgraph Security["Security Layer"]
            PERM[Fail-Closed Permissions]
            AUD[Audit Logger]
            INJ[Injection Detection]
        end

        API --> MAS
        API --> Compute
        API --> Tools
        API --> Security
    end

    Clients -->|HTTP + SSE| Runtime

Monorepo Structure

codein.pro/
├── packages/
│   ├── agent/              # Agent runtime — Node.js HTTP server (port 43120)
│   │   ├── src/
│   │   │   ├── index.js    #   Server bootstrap + 27 route modules
│   │   │   ├── mas/        #   Multi-Agent Swarm (13 agents, 4 topologies)
│   │   │   ├── routes/     #   HTTP route handlers
│   │   │   ├── ai-hub/     #   Multi-provider AI management
│   │   │   ├── mcp/        #   MCP client manager
│   │   │   ├── gpu-orchestration/  # RunPod integration
│   │   │   └── security/   #   Sandbox, validators, rate limiter
│   │   └── test/           #   193 test suites · 1,788 tests
│   │
│   ├── extension/          # VS Code / JetBrains adapter
│   ├── codein-sdk/         # SDK for programmatic agent integration
│   ├── config-yaml/        # YAML configuration parser
│   ├── config-types/       # Shared TypeScript config types
│   ├── compute-local/      # Local compute manager
│   ├── openai-adapters/    # OpenAI SDK adapters
│   └── fetch/              # Custom fetch wrapper with auth
│
├── core/                   # Shared TypeScript library
│   ├── llm/llms/           #   60 LLM provider implementations
│   ├── autocomplete/       #   Code completion engine
│   ├── context/            #   Context provider system
│   └── util/               #   Shared utilities
│
├── gui/                    # React 18 + Redux Toolkit + Vite 6
│   ├── src/
│   │   ├── components/     #   50+ components (panels, editor, terminal)
│   │   ├── pages/          #   Main GUI routes
│   │   ├── redux/          #   State slices (session, swarm, config, ui…)
│   │   └── util/           #   Auth, agent config, helpers
│   └── (68 test files · 641 tests)
│
├── electron-app/           # Electron 28 desktop shell
├── extensions/
│   ├── vscode/             # VS Code extension (40+ commands, 30+ keybindings)
│   ├── intellij/           # JetBrains IDE adapter
│   └── cli/                # Command-line interface
└── landing/                # Public website

Tech Stack

Layer	Technology
Desktop shell	Electron 28.2 with auto-update
Frontend	React 18, Redux Toolkit 2.11, Vite 6, Tailwind CSS 3.4
Code editor	Monaco Editor 4.7
Terminal	xterm.js 5.3 + node-pty
Rich input	TipTap 2.27
Agent runtime	Node.js 20.19+, plain `node:http` (no framework)
Persistence	better-sqlite3
Auth	jsonwebtoken 9.0
Logging	Pino
Platforms	Windows (NSIS), macOS (DMG/PKG), Linux (AppImage/DEB/RPM)

Security Model

CodeIn follows a defense-in-depth architecture for locally-run agent execution.

Layer	What it does
Fail-closed permissions	Every destructive operation requires explicit user approval before the agent proceeds
JWT authentication	All agent API requests require a valid token; configurable secret
Prompt injection detection	Pattern matching and structural heuristics block injected instructions
Input validation	Shell metacharacter blocking, path traversal prevention, schema validation
Docker sandbox	Agent code execution: dropped Linux capabilities, read-only root, CPU/memory limits, network isolation
Audit logging	Every agent action logged with timestamp, agent ID, and rationale to `data/audit-logs/`
Rate limiting	Sliding-window per-key rate limiter on all endpoints
Secret redaction	API keys automatically masked in logs and SSE streams
Zero telemetry	No tracking, no analytics, no phone-home of any kind

See SECURITY.md for the full model and responsible disclosure policy.

Getting Started

Prerequisites: Node.js 20.19+ and npm 10+

Install from source

# Clone the repository
git clone https://github.com/inbharat-ai/codein.pro.git
cd codein.pro
npm install

# Start the agent runtime (keep this running)
cd packages/agent
npm start
# → CodeIn Agent listening on http://127.0.0.1:43120

# In a second terminal — launch the desktop app (recommended)
cd ../..
cd electron-app && npm run dev

# Or launch the web GUI
cd gui && npm run dev

Verify the agent is running

curl http://127.0.0.1:43120/api/v1/health
# {"status":"ok","uptime":42,...}

Download pre-built binaries

Platform	Format	Link
Windows	NSIS installer + portable	Releases
macOS	DMG (Intel + Apple Silicon)	Releases
Linux	AppImage + DEB	Releases

VS Code Extension

Install the extension from packages/extension/ or build from source:

cd packages/extension
npm install && npm run build
# Load the generated .vsix in VS Code via "Install from VSIX"

Testing

Package	Runner	Suites	Tests
`packages/agent`	Node.js test runner	193	1,788
`gui`	Vitest + Testing Library	68	641
Total		261	2,429

cd packages/agent && npm test       # Agent runtime tests
cd gui && npx vitest run            # GUI component tests
cd gui && npx tsc --noEmit          # TypeScript type check

Project Status

Version: v1.0.3-beta | License: Apache 2.0 | Stage: Public beta

CodeIn's core architecture is complete and the primary feature set — multi-agent swarm, 60 provider integrations, local inference, and 22-language support — works end-to-end. The project is in active development toward v1.0 stable.

Implemented and working:

Multi-agent swarm (13 agents, 4 topologies, SQLite persistence)
60 LLM provider integrations with per-request cost tracking
Local offline inference via llama.cpp
22 Indian languages + code-mixed input + voice I/O
Docker sandboxed code execution
Git automation with approval gates
Repo intelligence (AST, semantic search, impact analysis)
Web research (search, fetch, cache)
MCP tool protocol (stdio + HTTP)
RunPod GPU orchestration
VS Code extension + JetBrains adapter + CLI
Electron desktop with auto-update
JWT authentication, rate limiting, audit logging

In progress:

Code Knowledge Graph (GitNexus) — AST-backed graph queries, Cypher interface, execution flow tracing
Computer Use vision pipeline — advanced screen-state understanding beyond current skill templates
Vibe Coding Mode — image-to-UI pipeline; intent parsing and spec generation functional, full code generation pipeline maturing

Known technical debt (being addressed):

packages/agent/src/index.js is monolithic (~1,100 lines); route modularization in progress
gui/src/redux/slices/sessionSlice is large; normalization planned
Mixed styling (Tailwind + styled-components + plain CSS); consolidating to Tailwind
No API versioning on route endpoints yet

Roadmap

Milestone	Target	Status
v1.0-beta public launch	Q1 2026	🟡 Now
GitNexus code graph GA	Q2 2026	🔄 In progress
Computer Use GA	Q2 2026	🔄 In progress
Vibe Coding GA	Q2 2026	🔄 In progress
VS Code Marketplace listing	Q2 2026	📋 Planned
JetBrains Marketplace listing	Q2 2026	📋 Planned
Route API versioning (`/api/v2/`)	Q2 2026	📋 Planned
sessionSlice normalization + perf	Q2 2026	📋 Planned
v1.0 stable release	Q3 2026	📋 Planned

Contributing

Contributions are welcome — bug fixes, new features, LLM provider integrations, language support expansions, documentation improvements, and test coverage.

# 1. Fork and clone
git clone https://github.com/YOUR_USERNAME/codein.pro.git

# 2. Create a feature branch
git checkout -b feat/your-feature

# 3. Make your changes and commit (conventional commits)
git commit -m 'feat(agent): add support for XYZ provider'

# 4. Push and open a pull request
git push origin feat/your-feature

Read CONTRIBUTING.md for conventions and DEVELOPMENT.md for the full developer setup guide.

License

Apache 2.0 — free to use, modify, and distribute, including commercially.

Contact

Built by inBharat AI · Email: info@inbharat.ai · Issues: github.com/inbharat-ai/codein.pro/issues

Your code. Your language. Your machine.

Name		Name	Last commit message	Last commit date
Latest commit History 21,368 Commits
.claude/skills		.claude/skills
.continue		.continue
.github		.github
.husky		.husky
.vscode		.vscode
actions		actions
binary		binary
core		core
docs		docs
electron-app		electron-app
eval		eval
extensions		extensions
gui		gui
landing		landing
manual-testing-sandbox		manual-testing-sandbox
media		media
packages		packages
scripts		scripts
skills		skills
sync		sync
.continueignore		.continueignore
.eslintrc.shared.json		.eslintrc.shared.json
.gitignore		.gitignore
.node-version		.node-version
.nvmrc		.nvmrc
.prettierignore		.prettierignore
.prettierrc		.prettierrc
ADVANCED_FEATURES_GUIDE.md		ADVANCED_FEATURES_GUIDE.md
ARCHITECTURE.md		ARCHITECTURE.md
BACKEND_API_REFERENCE.md		BACKEND_API_REFERENCE.md
CHANGELOG.md		CHANGELOG.md
CLA.md		CLA.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
COMPLETE_FEATURE_LIST.md		COMPLETE_FEATURE_LIST.md
COMPLETE_FEATURE_PARITY.md		COMPLETE_FEATURE_PARITY.md
COMPLETE_USAGE_GUIDE.md		COMPLETE_USAGE_GUIDE.md
CONTRIBUTING.md		CONTRIBUTING.md
DAILY_BUILD_SCHEDULE.md		DAILY_BUILD_SCHEDULE.md
DELIVERABLES.md		DELIVERABLES.md
DEVELOPMENT.md		DEVELOPMENT.md
DOCUMENTATION_INDEX.md		DOCUMENTATION_INDEX.md
Dockerfile		Dockerfile
FEATURES.md		FEATURES.md
HARDGATE_TEST_RESULTS_2026_02_28.md		HARDGATE_TEST_RESULTS_2026_02_28.md
LICENSE		LICENSE
LLM_BOOTSTRAP_SYSTEM.md		LLM_BOOTSTRAP_SYSTEM.md
PROJECT_SUMMARY.md		PROJECT_SUMMARY.md
QUICKREF.md		QUICKREF.md
QUICK_REFERENCE_CHEATSHEET.md		QUICK_REFERENCE_CHEATSHEET.md
README.md		README.md
README_DELIVERY.md		README_DELIVERY.md
README_START_HERE.md		README_START_HERE.md
SECURITY.md		SECURITY.md
SECURITY_AND_INTEGRATION.md		SECURITY_AND_INTEGRATION.md
SETUP_CODIN_ELITE.sh		SETUP_CODIN_ELITE.sh
SETUP_COMPLETE.md		SETUP_COMPLETE.md
START_HERE_NOW.md		START_HERE_NOW.md
TESTING.md		TESTING.md
VSIX_FIX_NOTES.md		VSIX_FIX_NOTES.md
build.ps1		build.ps1
docker-compose.yml		docker-compose.yml
package-lock.json		package-lock.json
package.json		package.json
setup-agent.bat		setup-agent.bat
test_agent.py		test_agent.py
tsconfig.json		tsconfig.json
worktree-config.yaml		worktree-config.yaml

Folders and files

Latest commit

History

Repository files navigation

CodeIn

What is CodeIn?

Why CodeIn?

Core Capabilities

AI Engine

Development Tools

Editor and UI

Multi-Agent Swarm

Language Support

Tier 1 — Full Support (11 languages)

Tier 2 — Core Support (11 languages)

Code-Mixed Input

LLM Providers

Skills and Automation

Agent Runtime Skills (12)

Claude Code IDE Skills (11)

Architecture

System Overview

Monorepo Structure

Tech Stack

Security Model

Getting Started

Install from source

Verify the agent is running

Download pre-built binaries

VS Code Extension

Testing

Project Status

Roadmap

Contributing

License

Contact

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages