Autonomous Intelligent Development Environment
An AI agent framework that reads, plans, edits, tests, and verifies code changes autonomously.
aide is a Python framework for building autonomous coding agents — AI systems that can understand a codebase, plan multi-step tasks, implement changes, run tests, and self-verify the results.
Unlike single-prompt code generators, aide agents operate in a plan → execute → verify loop, can recover from errors, and collaborate across multiple specialized roles.
pip install aide-agent
from aide import Agent, Codebase agent = Agent(model="gpt-4o") repo = Codebase("./my-project") # Ask the agent to implement a feature result = agent.run( "Add rate limiting to the /api/login endpoint using Redis. " "Include unit tests and update the API docs.", repo=repo, ) print(result.summary) print(result.files_changed) print(result.tests_passed) # True/False
Or use the CLI:
# Analyze a codebase aide scan ./my-project # Implement a feature with full autonomy aide run ./my-project --task "Add input validation to all REST endpoints" # Review uncommitted changes aide review --staged # OpenAI Codex integration aide codex exec ./my-project --task "Fix the failing tests in auth module"
| Feature | Description |
|---|---|
| Plan-Execute-Verify Loop | Agents plan multi-step tasks, execute them, and verify results before proceeding |
| Multi-Agent Orchestration | Specialized agents (Planner, Coder, Reviewer, Tester) collaborate on complex tasks |
| Code Intelligence | AST-aware editing, symbol resolution, import management, type inference |
| Git Integration | Branch, commit, diff, cherry-pick — agents work within your Git workflow |
| GitHub Automation | PR review, issue triage, automated fixes pushed to branches |
| Self-Healing | Agents detect test failures and iteratively fix their own code |
| Tool Registry | Composable tools — file I/O, shell, search, git, test runners |
| Streaming & Async | Real-time output with async support for concurrent agent tasks |
| Checkpoint & Resume | Save agent state mid-task; resume from any checkpoint |
| Human-in-the-Loop | Optional approval gates for high-risk operations |
┌─────────────────┐
│ User / CLI │
└────────┬────────┘
│
┌────────▼────────┐
│ Orchestrator │
│ (task router) │
└────────┬────────┘
┌─────────────┼─────────────┐
│ │ │
┌────────▼───┐ ┌──────▼──────┐ ┌────▼────────┐
│ Planner │ │ Coder │ │ Reviewer │
│ Agent │ │ Agent │ │ Agent │
└────────┬────┘ └──────┬──────┘ └────┬────────┘
│ │ │
└─────────────┼─────────────┘
│
┌────────▼────────┐
│ Tool Layer │
├─────────────────┤
│ fs │ git │ shell│
│ search │ tests │
└─────────────────┘
from aide import Agent, Codebase agent = Agent(model="gpt-4o") repo = Codebase("./project") result = agent.run( "The login endpoint returns 500 when the email field is missing. " "Fix the bug, add proper error handling, and write a regression test.", repo=repo, verify=True, # Agent will run tests after fixing ) # Result includes: # - What files were changed and why # - Test results (before and after) # - A summary suitable for a commit message
from aide import GitHubAgent reviewer = GitHubAgent(model="gpt-4o") # Review a PR and post comments reviewer.review_pr( repo="owner/repo", pr_number=42, focus=["security", "performance"], post_comments=True, )
from aide import Pipeline, PlannerAgent, CoderAgent, ReviewerAgent pipeline = Pipeline([ PlannerAgent(), # Break task into subtasks CoderAgent(), # Implement each subtask ReviewerAgent(), # Review and suggest improvements ]) result = pipeline.run( "Refactor the database layer to use the repository pattern", repo=Codebase("./project"), )
Modern AI code tools are either single-shot (one prompt, one response) or workflow-locked (fixed pipelines). aide provides a composable framework where agents can be mixed, extended, and customized for your specific development workflow.
Built for developers who want AI that thinks in steps, not just completes lines.
# aide.yaml model: gpt-4o agents: planner: temperature: 0.2 coder: temperature: 0.1 max_retries: 3 reviewer: temperature: 0.3 tools: - fs # File system operations - git # Git commands - shell # Shell execution - search # Code search (ripgrep-based) - test_runner # Auto-detect and run tests verify: auto_test: true test_command: "pytest -x" lint_command: "ruff check ."
-
Install:
pip install aide-agent
-
Set your API key:
export OPENAI_API_KEY=sk-xxx -
Run:
aide run ./your-project --task "Add input validation to the registration form"
We welcome contributions! See CONTRIBUTING.md for guidelines.
Apache 2.0 — see LICENSE.