Overnight SWE

Recommended Stacks

One-click to select a curated combination of tools

Or build your own stack

Mix and match tools from the categories below. Click a tool to add it to your stack.

Multi-Agent Frameworks

Orchestrate multiple AI agents with specialized roles

MetaGPT

Multi-agent framework with PM, Architect, Engineer, QA roles. Simulates a full software company.

open-sourcepythonmature

Claude-Flow

64-agent orchestration for Claude Code. 84.8% SWE-Bench solve rate. MCP protocol native.

claude-nativetypescriptenterprise

CrewAI

43k stars. Role-based AI agent teams. Most popular multi-agent framework.

43k-starspythonpopular

AG2 (AutoGen)

37k stars. Microsoft's multi-agent conversation framework. Human-in-the-loop.

37k-starsmicrosoftconversational

LangGraph

24k stars. Fastest framework, lowest latency. Graph-based state machines.

24k-starsfastgraph-based

Letta

21k stars. Stateful agents with self-editing memory. Agents learn and improve over time.

21k-starsstatefulself-improving

Pydantic AI

14k stars. Type-safe Python agent framework. FastAPI-like DX for AI agents.

14k-starstype-safepython

AWS Agent Squad

AWS's open-source multi-agent orchestrator. Flexible routing and coordination.

awsopen-sourcerouting

ALMAS

Academic framework following agile SDLC. Modular design for human-agent collaboration.

researchagilemodular

AI Coding Agents

Autonomous agents that write, edit, and debug code

Claude Code YOLO

Claude Code with --dangerously-skip-permissions. Full autonomous mode, zero confirmations.

officialautonomousrecommended

Cline

57k stars. #1 open-source coding agent. VS Code native, 4M+ developers. Plan/Act modes.

57k-starsvscoderecommended

Aider

40k stars. Terminal-based AI pair programming. Auto git commits, multi-file edits.

40k-starsterminalgit-native

OpenCode

70k stars. Open source terminal agent. 650k+ developers, multi-provider support.

70k-starsterminalpopular

Roo Code

901k installs. Fork of Cline with multi-agent modes. Code/Architect/Debug roles.

cline-forkmulti-modevscode

Continue

20k stars. Open-source Copilot alternative. Agent, Chat, Edit, Autocomplete modes.

20k-starsvscodeautocomplete

Devin

Cognition's autonomous SWE. Full dev environment with terminal, editor, browser.

commercialautonomousfull-stack

SWE-Agent

Open-source Devin alternative. Takes GitHub issues, fixes code end-to-end via Docker.

open-sourcegithub-issuesdocker

Devon

Lightweight open-source SWE. Plans, edits, tests, debugs. Runs locally.

open-sourcelightweightlocal

OpenHands

Research framework for multi-agent experimentation. SWE-bench evaluation harness.

researchbenchmarksevaluation

Orchestrators

Schedule and coordinate agent workflows

Mastra

TypeScript AI framework from Gatsby team. Graph workflows, human-in-loop, 40+ providers.

typescriptlocalrecommended

Open Agent

Self-hosted orchestrator for Claude Code & OpenCode. Isolated Linux workspaces.

self-hostedclaude-nativeisolated

Windmill

Fastest workflow engine (13x vs Airflow). Self-hostable Temporal alternative.

self-hostedfastopen-source

n8n

Visual workflow automation. Self-hosted, free tier. Great for AI agent orchestration.

visualself-hostedpopular

Activepieces

400+ MCP servers for AI agents. Open-source Zapier alternative. AI-native design.

mcpopen-sourceai-native

Temporal

Mission-critical durable execution. Handles failures gracefully. Open-source.

durableenterprisereliable

Inngest

Event-driven durable functions. TypeScript-first with full type inference.

event-driventypescripttyped

Trigger.dev

Background jobs platform. Native Mastra integration for agent workflows.

hostedtypescriptmastra-compatible

Kestra

Declarative YAML workflows. Event-driven, multi-language. Infrastructure as Code.

yamlevent-driveniac

SDKs & APIs

Build custom agents programmatically

Claude Agent SDK

Official Anthropic SDK for building Claude Code agents. Python & TypeScript.

officialpythontypescript

Claude Code GitHub Actions

Turn GitHub issues into PRs automatically. Official Anthropic integration.

officialgithubci-cd

OpenAI Agents SDK

18k stars. Official OpenAI SDK for building agents. Function calling, threads.

18k-starsofficialopenai

Google ADK

17k stars. Google's Agent Development Kit. Gemini-native, Python SDK.

17k-starsgooglegemini

Agent Memory

Persistent memory and learning for AI agents

Mem0

37k stars. Universal memory layer for AI agents. +26% accuracy, 90% token savings.

37k-starsrecommendedproduction

Letta Memory SDK

Self-editing memory for LLMs. Agents update their own memory to learn over time.

self-editinglearningstateful

LangGraph + MongoDB

Long-term memory for LangGraph agents. Cross-session persistence with MongoDB.

langgraphmongodbpersistent

Cloud Compute

Cheap/free compute for overnight runs

E2B

AI agent sandboxes. ~$0.05/hr, per-second billing, starts in <200ms.

cheapfastsandboxed

Modal

AI infrastructure that scales to zero. $30/mo free credits included.

free-tierserverlessbatch

Daytona

Secure platform for AI-generated code. $200 free credits, ~$0.067/hr.

free-tiersecureai-focused

RunPod

GPU cloud with per-second billing. Community Cloud for cheap R&D.

gpucheapcommunity

Vast.ai

GPU marketplace. 50-70% cheaper than hyperscalers. Bid for spare capacity.

marketplacecheapbidding

TensorDock

Starting at $0.12/hr for consumer GPUs. Global coverage, marketplace model.

cheapmarketplaceglobal

Thunder Compute

$0.66/hr for A100 40GB. True pay-as-you-go, no commitments.

a100pay-as-you-gono-commitment

Beam Cloud

Ultra-fast cold starts (2-3s). Python SDK with hot-reloading for iterative dev.

fast-cold-startpythoniterative

GCP Preemptible

60-90% discount on Google Cloud. Perfect for interruptible overnight jobs.

cheapinterruptiblegoogle

Azure Spot VMs

65-69% discount on Azure. Great for batch workloads.

cheapinterruptiblemicrosoft

Cron & Schedulers

Trigger agents on a schedule

Supabase pg_cron

PostgreSQL-native cron jobs. Free tier available, triggers Edge Functions.

free-tierpostgresedge-functions

Vercel Cron

Serverless cron jobs. Pro plan ($20/mo) has 60s timeout - good for triggers.

serverless60s-limittrigger-only

GitHub Actions Cron

Free cron for public repos. 6-hour timeout. Good for overnight runs.

freegithublong-running

Agent Roles

Specialized agent personas for your team

Product Manager

Orchestrator agent. Reviews specs, manages team, approves PRs, updates roadmap.

orchestratorreviewsspecs

QA Engineer

Writes tests, validates against specs, reports failures, blocks bad PRs.

testingvalidationquality

SWE Agent

Implements features from specs. Creates commits/PRs. Follows coding standards.

codingimplementationmultiple

Architect

Designs technical solutions. Reviews architecture decisions. Maintains system design.

designarchitecturereview

Project Manager

Breaks down tasks, manages dependencies, tracks progress, unblocks team.

tasksdependenciestracking

Development Approaches

Methodologies for autonomous development

Spec-Driven Development

Write specs first, then implement. 1 iteration with specs = 8 without.

specs-firstreliablerecommended

Agentic SDLC

Agents follow traditional SDLC phases: requirements, design, implement, test, deploy.

traditionalphasedstructured

Shared Thread System

Agents post updates to shared thread. Hourly standups. Self-awareness loops.

communicationstandupsself-aware