Skip to content

BigOD2307/AutoDev

BranchesTags

Repository files navigation


     _         _        ____
    / \  _   _| |_ ___ |  _ \  _____   __
   / _ \| | | | __/ _ \| | | |/ _ \ \ / /
  / ___ \ |_| | || (_) | |_| |  __/\ V /
 /_/   \_\__,_|\__\___/|____/ \___| \_/
                                    v2.0

The autonomous AI agent that continuously improves your codebase.

Security. Performance. SEO. Content. AI Quality.
All on autopilot. Zero risk.

Quick Start • How It Works • Smart Router • Modules • Memory • Dashboard • Deploy

The Problem

Your codebase has dozens of improvements waiting to happen. Missing meta tags. Unvalidated API inputs. Static imports that should be lazy-loaded. Prompts that could be sharper. You know they're there. You just don't have time.

The Solution

AutoDev runs every 4 hours. It picks one improvement, applies it, verifies the build passes, and creates a branch for your review. If anything breaks, it automatically reverts. You merge when ready.

  You                                     AutoDev
  ----                                    -------
  Sleeping                                Analyzing 38 API routes...
  Sleeping                                Found: /api/auth missing rate limiting
  Sleeping                                Generated fix with DeepSeek V3.2
  Sleeping                                Build passed in 16s
  Sleeping                                Pushed branch: autodev/security/2026-03-24
  Wake up                                 Sent notification: "New improvement ready"
  Review diff on GitHub
  Click merge

Why AutoDev?

Feature AutoDev Manual code reviews Linters
Finds missing auth on routes Yes Sometimes No
Generates blog content Yes No No
Fixes performance anti-patterns Yes Sometimes Partially
Adds JSON-LD structured data Yes No No
Improves AI prompts Yes No No
Learns from past mistakes Yes No No
Runs while you sleep Yes No Yes
Zero risk to production Yes Depends Yes

Quick Start

30-second setup (single project)

git clone https://github.com/BigOD2307/AutoDev.git
cd AutoDev && npm install
export OPENROUTER_API_KEY=sk-or-v1-xxxx    # Get at openrouter.ai/keys
export GITHUB_TOKEN=ghp_xxxx               # github.com/settings/tokens (scope: repo)
export AUTODEV_REPO_URL=https://github.com/you/your-project.git
export AUTODEV_PROJECT_NAME=my-project
export AUTODEV_DRY_RUN=true                # Safe mode first

npx tsx src/index.ts

Open http://localhost:4040 to see the dashboard.

Multi-project setup

cp autodev.config.example.json autodev.config.json

Edit autodev.config.json:

{
  "projects": [
    {
      "name": "my-saas",
      "repo": "https://github.com/you/my-saas.git",
      "framework": "nextjs",
      "modules": ["security", "performance", "seo", "quality"],
      "keyFiles": ["app/layout.tsx", "middleware.ts", "lib/auth.ts"]
    },
    {
      "name": "my-api",
      "repo": "https://github.com/you/my-api.git",
      "framework": "express",
      "modules": ["security", "performance"],
      "keyFiles": ["src/index.ts", "src/middleware.ts"]
    }
  ]
}
npm run dry-run

How It Works

                         Every 4 hours
                              |
                    +---------v---------+
                    |  Pick next module  |  Intelligent rotation
                    |  Check memory for  |  Skip failing modules
                    |  fragile files     |  Reuse winning strategies
                    +---------+---------+
                              |
                    +---------v---------+
                    |  git pull latest   |
                    |  npm install       |
                    +---------+---------+
                              |
                    +---------v---------+
                    |  Smart Router      |  Picks best model:
                    |  scores 10 models  |  DeepSeek, Qwen, Gemini,
                    |  picks top 3       |  MiniMax, Llama, Claude...
                    +---------+---------+
                              |
                    +---------v---------+
                    |  LLM analyzes code |  Reads key files
                    |  + memory context  |  Knows past errors
                    |  Proposes ONE fix  |  Knows fragile files
                    +---------+---------+
                              |
                    +---------v---------+
                    |  Apply changes     |
                    |  Run build         |
                    +---------+---------+
                              |
                     +--------+--------+
                     |                 |
                  PASSED            FAILED
                     |                 |
              +------v------+   +------v------+
              | git branch  |   | git revert  |
              | git commit  |   | record error|
              | git push    |   | learn from  |
              | notify team |   | the mistake |
              +-------------+   +-------------+

Smart LLM Router

AutoDev doesn't depend on one model. It has a smart router that picks the best model for each task, optimizing cost and quality.

10 Models, Automatic Selection

Task: security fix
  Score algorithm:
    Quality x CostEfficiency x HistoricalSuccessRate x SpeedBonus x StrengthMatch

  Results:
    #1  DeepSeek V3.2    Score: 495   $0.28/1M out   SELECTED
    #2  Qwen 3 235B      Score: 348   $0.30/1M out   Fallback 1
    #3  Llama 4 Maverick  Score: 217   $0.60/1M out   Fallback 2
Model $/1M out Quality Speed Best for
DeepSeek V3.2 $0.28 92 Fast Code, Security
Qwen 3 235B $0.30 90 Medium Code, Analysis
Gemini 2.5 Flash $0.60 88 Fast Content, SEO
MiniMax M1 $1.10 87 Medium Code, Content
Llama 4 Maverick $0.60 85 Fast General
Mistral Medium 3 $2.00 88 Medium Code, Content
GPT-4o Mini $0.60 83 Fast General
Claude Haiku 4.5 $4.00 91 Fast Code, Eval
DeepSeek R1 $2.19 95 Slow Deep Analysis
Gemini 2.5 Pro $10.00 94 Slow Deep Eval

Automatic fallback: if model #1 times out, the router tries #2, then #3.

Self-learning: the router tracks success/failure per model and adjusts scores over time. Bad models get deprioritized automatically.

Monthly cost: ~$1-2 with default settings.

Modules

Security

Detects and fixes vulnerabilities across your API routes:

Scans for:
  POST routes without authentication        [HIGH]
  JSON body parsed without validation        [MEDIUM]
  Server env vars in client-side code        [CRITICAL]
  dangerouslySetInnerHTML (XSS risk)         [MEDIUM]
  Missing rate limiting on auth endpoints    [MEDIUM]

Performance

Finds and fixes performance anti-patterns:

Detects:
  framer-motion imported statically          -> dynamic import
  <img> without optimization                 -> next/image or equivalent
  5+ sequential awaits                       -> Promise.all
  Dynamic imports without loading states     -> add Suspense/loading

SEO

Adds missing SEO elements:

Checks:
  generateMetadata() on every page
  Open Graph tags (og:title, og:description, og:image)
  JSON-LD structured data
  Sitemap.xml and robots.txt
  Canonical URLs and hreflang tags

Content

Generates blog articles for your audience:

Configure in autodev.config.json:
  "content": {
    "enabled": true,
    "blogFile": "lib/blog.ts",
    "topics": ["ai-tools", "productivity", "engineering"],
    "targetAudience": "developers",
    "language": "en"
  }

The agent reads your existing blog format, avoids duplicates, researches trending topics via web search, and generates articles matching your style.

Quality

Improves AI system prompts in your codebase:

Detects system prompts automatically (SYSTEM_PROMPT, systemPrompt, etc.)
Evaluates: clarity, completeness, actionability, consistency
Suggests targeted improvements
Never changes output format or business logic

Memory & Learning

AutoDev has a persistent memory that makes it smarter over time. This isn't just logging — it's active learning that changes the agent's behavior.

What it remembers

data/memory.json
{
  "projects": {
    "my-saas": {
      "totalRuns": 47,
      "modules": {
        "security": {
          "totalAttempts": 12,
          "totalSuccesses": 9,
          "totalFailures": 3,
          "consecutiveFailures": 0,
          "errorPatterns": ["error TS2345", "Module not found"],
          "fragileFiles": ["lib/auth.ts"],
          "successfulStrategies": [
            "Added zod validation to POST route",
            "Rate limiting with sliding window"
          ]
        }
      },
      "fileRisk": {
        "lib/auth.ts": 4,          // HIGH risk — handle with care
        "app/page.tsx": 1,          // LOW risk — safe to modify
        "middleware.ts": 2           // MEDIUM risk
      }
    }
  }
}

How it learns

Event Memory action
Build passes File risk -1, strategy recorded
Build fails File risk +2, error pattern saved
Risk >= 4 File marked as "fragile"
3+ consecutive failures Module skipped next run

Memory is injected into LLM prompts

The LLM receives past context before generating:

## AGENT MEMORY (learn from past runs)
Success rate: 75% (9/12)

Errors to AVOID:
- error TS2345 (type mismatch)
- Module not found

Fragile files (modify with EXTRA CARE):
- lib/auth.ts (risk: 4)

Strategies that WORKED before:
- Added zod validation to POST route

Dashboard

AutoDev includes a built-in web dashboard at http://localhost:4040.

+------------------------------------------------------------------+
|  AutoDev Dashboard                              [Idle] [Run Now]  |
+------------------------------------------------------------------+
|                                                                   |
|  Total Runs    Success Rate    Successes    Failed                 |
|     47            75%             35           12                  |
|                                                                   |
|  +------------------+  +-------------------------------------+   |
|  | Projects         |  | Recent Improvements                 |   |
|  |                  |  |                                     |   |
|  | my-saas          |  | [success] security  Rate limiting   |   |
|  |   nextjs         |  | [failed]  perf      Lazy loading    |   |
|  |   4 modules      |  | [success] seo       JSON-LD added   |   |
|  |                  |  | [success] quality   Prompt improved  |   |
|  +------------------+  +-------------------------------------+   |
|  | Agent Memory     |  | Live Logs                            |   |
|  |                  |  |                                     |   |
|  | security 75%     |  | [INFO] Pulling latest for my-saas   |   |
|  | perf     60%     |  | [INFO] Smart Router: DeepSeek V3.2  |   |
|  | seo      90%     |  | [SUCCESS] Found: Add JSON-LD...     |   |
|  | Fragile:auth.ts  |  | [INFO] Build passed (12s)            |   |
|  +------------------+  +-------------------------------------+   |
+------------------------------------------------------------------+

Features:

  • Real-time logs via Server-Sent Events (SSE)
  • Agent status with live status indicator
  • Run Now button for on-demand runs
  • Memory visualization — see success rates, fragile files, error patterns
  • Project overview — configured projects and their modules
  • Improvement history with status, module, timestamp

Access it at http://localhost:4040 (configurable via DASHBOARD_PORT).

Deploy

Railway (Recommended)

$20 free credit with our referral: railway.com?referralCode=oPjPCV

  1. Fork this repo
  2. Railway -> New Project -> Deploy from GitHub
  3. Add environment variables
  4. Done. The Dockerfile handles everything.

Docker

docker build -t autodev .
docker run --env-file .env -p 4040:4040 autodev

PM2

npm run build
pm2 start dist/index.js --name autodev

Any VPS

git clone https://github.com/BigOD2307/AutoDev.git
cd AutoDev && npm install && npm run build
node dist/index.js

Safety Guarantees

Layer What it does
Git isolation Never pushes to main. Always a separate branch.
Build verification Full build after every change. Fails? Auto-revert.
Auto-revert git checkout -- . + git clean -fd on any failure.
Human review You review every diff. You click merge. Always.
Memory Learns from failures. Skips fragile files. Avoids known errors.
Dry run mode Test everything without pushing. AUTODEV_DRY_RUN=true

Supported Frameworks

Framework Modules available
Next.js All 5 (security, perf, seo, content, quality)
React / Vite Security, Performance, Quality
Vue / Nuxt Security, Performance, SEO, Quality
Svelte / SvelteKit Security, Performance, SEO, Quality
Express / Fastify Security, Performance, Quality
Django / Flask Security, Quality
Any Node.js project Security, Performance, Quality

Configuration Reference

autodev.config.json

{
  "agent": {
    "name": "AutoDev",                    // Agent display name
    "schedule": "0 */4 * * *",            // Cron schedule
    "dryRun": false                       // true = no push, no notifications
  },
  "projects": [{
    "name": "my-project",                 // Project identifier
    "repo": "https://github.com/...",     // Git clone URL
    "branch": "main",                     // Target branch
    "buildCommand": "npm run build",      // Build verification command
    "framework": "nextjs",                // Framework hint for the LLM
    "modules": ["security", "performance", "seo", "quality"],
    "keyFiles": ["app/layout.tsx", "..."],// Files the LLM always reads
    "paths": {                            // Project structure hints
      "pages": "app/",
      "api": "app/api/",
      "config": "next.config.js"
    },
    "content": {                          // Blog generation config
      "enabled": false,
      "blogFile": "lib/blog.ts",
      "topics": ["topic-1", "topic-2"],
      "targetAudience": "developers",
      "language": "en"
    }
  }],
  "notifications": {
    "channels": [
      { "type": "slack", "url": "https://hooks.slack.com/..." },
      { "type": "discord", "url": "https://discord.com/api/webhooks/..." },
      { "type": "webhook", "url": "https://...", "secret": "..." }
    ]
  },
  "git": {
    "authorName": "AutoDev Agent",
    "authorEmail": "autodev@company.com",
    "branchPrefix": "autodev"
  }
}

Environment Variables

Variable Required Default
OPENROUTER_API_KEY Yes —
GITHUB_TOKEN Yes —
AUTODEV_DRY_RUN No false
AUTODEV_RUN_NOW No false
AUTODEV_SCHEDULE No 0 */4 * * *
DASHBOARD_PORT No 4040
NOTIFY_WEBHOOK_URL No —
AUTODEV_SECRET No —
TINYFISH_API_KEY No —

Contributing

  1. Fork the repo
  2. Create a feature branch
  3. Make your changes
  4. Open a PR

Add a new module

// src/modules/mymodule.ts
import { analyzeForImprovement, type Improvement } from '../core/analyzer.js'
import type { RepoConfig } from '../config.js'

export async function runMyModule(repo: RepoConfig): Promise<Improvement | null> {
  // Your pre-analysis logic here
  return analyzeForImprovement(repo, 'mymodule')
}

Register in src/core/agent.ts:

import { runMyModule } from '../modules/mymodule.js'
MODULE_RUNNERS['mymodule'] = runMyModule

Add a new LLM model

Edit src/utils/router.ts:

{
  id: 'provider/model-name',
  name: 'Display Name',
  costIn: 0.10,        // $/1M input tokens
  costOut: 0.50,        // $/1M output tokens
  contextWindow: 128000,
  strengths: ['code', 'analysis'],
  quality: 85,          // 0-100
  speed: 'fast',        // fast | medium | slow
}

License

Free to use for personal and internal projects. Attribution required:

Powered by AutoDev by Dicken AI

Commercial use (resale, SaaS, white-labeling) requires a license. Contact: ousmane@dickenai.com

See LICENSE.md for full terms.


AutoDev by Dicken AI
Building AI for Africa and the world.

GitHub • Contact • Deploy on Railway ($20 free)

About

The autonomous AI agent that continuously improves your codebase.

Topics

code autonomous-agents ai-agents ai-assistant

Resources

Readme

License

View license
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages