The coding agent built for long-horizon tasks
It plans before it codes, remembers across sessions, and runs for hours at a fraction of the cost.
Trusted by startups from
Terminal-Bench 2.0 · open dataSmaller model. Better performance.
Smaller model. Better performance.
3× less cost.
We outscore Claude Code and OpenCode on Terminal-Bench 2.0 — a full model tier down, at a third of the cost.
Beating Sonnet with Haiku
pre.dev + Haiku 4.5 outscores Claude Code at both the Haiku and Sonnet 4.5 tiers
Terminal-Bench 2.0 · pass@1 · n=89
Beating Opus with Sonnet
pre.dev + Sonnet 4.6 outscores Claude Code and OpenCode on Opus 4.5 — one tier down, still ahead
Terminal-Bench 2.0 · pass@1 · n=89
pre.dev runs the marathon, not the sprint
It decomposes the work, executes each step in isolated context, and verifies before moving on — for as long as it takes.
Describe the task
It plans and decomposes the build
Agents execute, verify, and open PRs
|
Describe the task
It plans and decomposes the build
Agents execute, verify, and open PRs
|
Built for work that won't fit in one context window
Most agents lose the plot on long tasks. pre.dev plans up front, holds state in persistent memory, and verifies every step.
Plans before it codes
Before writing a line, pre.dev researches the goal and decomposes the full build—milestones, stories, subtasks. Then executes it autonomously.
v1
14.29%
Tasks
Hours
Progress
10
11
12
13
14
15
16
17
18
19
20
21
M1: Milestone 1: ...
8h
100%
M1: Milestone 1: .
Write concise d...
8h
100%
M2: Milestone 2: ...
26h
20%
M2: Milestone 2: .
Hero Section Di...
2h
0%
Download/Signup...
4h
0%
Feature Benefit...
4h
0%
Signup Form
8h
0%
Password Reset ...
8h
0%
M3: Milestone 3: ...
32h
0%
M3: Milestone 3: .
One run ships the whole roadmap
Kick off a build and focused agents tackle each task in isolated context. No prompting, no babysitting—just verified code landing in your repo.
>Build Next Task
>Build On Autopilot
>Chat
ProPressEnterto send
Every task verified before moving on
The agent can't mark a task complete until it passes type checks, linting, and visual browser verification. No more 'looks good but is broken.'
Acceptance Verification
Criterion
Status
Hero section renders with gradient background and CTA redirects to /signup
Passed
TypeScript compilation passes with no errors and all unit tests succeed
Passed
API response under 200ms and error handling shows user-friendly messages
Passed
PR-native, never touches main
Every completed task opens a feature branch and PR. The agent never pushes to main. You review and merge on your terms.
Deployment+
Download
feat/dtc-913→predev·631a912
PR #1
feat(Setup): Write concise design aesthetic/rules
PR: ## Node: Setup · Show more
</> Code⎇ Diff
v2📋workspace/package.json
+251+1"private": true,
2"name": "mobile",
3"sideEffects": false,
4"scripts": {
+5 "build": "react-router build",
6 "clean": "node scripts/clean.js",
+7 "dev": "react-router dev",
8 "test": "vitest run",
Bring your codebase, it extends it
Reverse-engineers existing repos, adapts to any stack, and keeps building—no starting from scratch.
Search repositories...
stripe-syncPrivate
Updated today · Created 3mo agonexus-apiPrivate
Updated 2d ago · Created 1y agoquantum-ui
Updated today · Created 6mo agoforge-cliPrivate
Updated 1w ago · Created 2y agopulse-analyticsPrivate
Updated 3d ago · Created 4mo agoterraform-aws
Updated today · Created 8mo agoRefactor or build new features
Import
Your keys. Anything you can build against.
Connect your services once. Agents use them automatically while they build.
Recommendations
Exa
Neural search with Exa Fast (<350ms), Exa Auto for intelligent routing, Exa Deep for agentic multi-step retrieval, and /research endpoint for automated in-depth analysis with citations
EXA_API_KEY
•••••••••••••
Tavily
Real-time search API with Search, Extract, Map, and Crawl endpoints optimized for RAG. SOC-2 certified, trusted by 800K+ developers, with MCP server support
TAVILY_API_KEY
•••••••••••••
Perplexity
Sonar models with web-grounded LLM responses at 1200 tokens/sec. Features search modes (High/Medium/Low), Pro Search with multi-step reasoning, and Deep Research for complex analysis
PERPLEXITY_API_KEY
•••••••••••••
Brave Search
Independent 30B+ page index with AI Grounding for RAG, rich snippets, Goggles for result re-ranking, and multi-endpoint support (web, images, news, local). Powers Claude MCP and major AI apps
BRAVE_SEARCH_API_KEY
•••••••••••••
You.com
Web Search API with Express (search+answer in one call), Contents API for page extraction, and News/Image APIs. 99.9% uptime, enterprise-grade with Databricks and AWS integrations
YOU_API_KEY
•••••••••••••
Firecrawl
Web-to-markdown with /scrape, /crawl, /map, /search, and /extract endpoints. Features /agent for complex navigation, natural language prompts (no CSS selectors), and 98% extraction accuracy
FIRECRAWL_API_KEY
•••••••••••••
Serper
Ultra-fast Google Search API with 1-2 second responses, 300 queries/sec, web/images/news/videos/maps endpoints, and MCP server support. Pricing from $0.30/1K queries
SERPER_API_KEY
•••••••••••••
SerpAPI
Scrape 80+ search engines with Google AI Mode API for GEO optimization, AI Overviews support, Travel Explore API, and MCP server for AI agents. Auto CAPTCHA solving and proxy rotation
SERPAPI_API_KEY
•••••••••••••
Browserbase
Serverless browser infrastructure with Stagehand v3 SDK (act/extract/observe/agent). 44% faster, multi-language support, self-healing automation, session replay, and CAPTCHA solving
BROWSERBASE_API_KEY
•••••••••••••
Apify
10,000+ pre-built Actors for any website, Crawlee library for custom crawlers, MCP server for AI agents, and integrations with LangChain/LlamaIndex/vector DBs. 48K+ GitHub stars
APIFY_API_TOKEN
•••••••••••••
Jina AI
Reader API (r.jina.ai) for URL-to-markdown with ReaderLM-v2, Search API (s.jina.ai), jina-embeddings-v4 (multimodal 30+ languages), and jina-reranker-v3. Single key for all services
JINA_API_KEY
•••••••••••••
ScrapingBee
Web scraping API with headless browser rendering, automatic proxy rotation, AI-powered natural language extraction, CAPTCHA solving, and Stealth Proxy (beta) for hard-to-access sites
SCRAPINGBEE_API_KEY
•••••••••••••
Exa
Neural search with Exa Fast (<350ms), Exa Auto for intelligent routing, Exa Deep for agentic multi-step retrieval, and /research endpoint for automated in-depth analysis with citations
EXA_API_KEY
•••••••••••••
Tavily
Real-time search API with Search, Extract, Map, and Crawl endpoints optimized for RAG. SOC-2 certified, trusted by 800K+ developers, with MCP server support
TAVILY_API_KEY
•••••••••••••
Perplexity
Sonar models with web-grounded LLM responses at 1200 tokens/sec. Features search modes (High/Medium/Low), Pro Search with multi-step reasoning, and Deep Research for complex analysis
PERPLEXITY_API_KEY
•••••••••••••
Brave Search
Independent 30B+ page index with AI Grounding for RAG, rich snippets, Goggles for result re-ranking, and multi-endpoint support (web, images, news, local). Powers Claude MCP and major AI apps
BRAVE_SEARCH_API_KEY
•••••••••••••
You.com
Web Search API with Express (search+answer in one call), Contents API for page extraction, and News/Image APIs. 99.9% uptime, enterprise-grade with Databricks and AWS integrations
YOU_API_KEY
•••••••••••••
Firecrawl
Web-to-markdown with /scrape, /crawl, /map, /search, and /extract endpoints. Features /agent for complex navigation, natural language prompts (no CSS selectors), and 98% extraction accuracy
FIRECRAWL_API_KEY
•••••••••••••
Serper
Ultra-fast Google Search API with 1-2 second responses, 300 queries/sec, web/images/news/videos/maps endpoints, and MCP server support. Pricing from $0.30/1K queries
SERPER_API_KEY
•••••••••••••
SerpAPI
Scrape 80+ search engines with Google AI Mode API for GEO optimization, AI Overviews support, Travel Explore API, and MCP server for AI agents. Auto CAPTCHA solving and proxy rotation
SERPAPI_API_KEY
•••••••••••••
Browserbase
Serverless browser infrastructure with Stagehand v3 SDK (act/extract/observe/agent). 44% faster, multi-language support, self-healing automation, session replay, and CAPTCHA solving
BROWSERBASE_API_KEY
•••••••••••••
Apify
10,000+ pre-built Actors for any website, Crawlee library for custom crawlers, MCP server for AI agents, and integrations with LangChain/LlamaIndex/vector DBs. 48K+ GitHub stars
APIFY_API_TOKEN
•••••••••••••
Jina AI
Reader API (r.jina.ai) for URL-to-markdown with ReaderLM-v2, Search API (s.jina.ai), jina-embeddings-v4 (multimodal 30+ languages), and jina-reranker-v3. Single key for all services
JINA_API_KEY
•••••••••••••
ScrapingBee
Web scraping API with headless browser rendering, automatic proxy rotation, AI-powered natural language extraction, CAPTCHA solving, and Stealth Proxy (beta) for hard-to-access sites
SCRAPINGBEE_API_KEY
•••••••••••••
Long-horizon execution
Agents run for hours, days, or weeks—holding state across the entire task instead of losing it mid-build.
Secure sandbox environment
Your code runs in an isolated environment. Build with confidence—zero risk to production.
Adaptive tech stack
Any framework, any language. The agent adapts to your stack, not the other way around.
Multi-service architectures
Model any combination of services and their interactions in a single sandbox environment. Frontend, backend, database, cache, and more.
Start building code that scales past prototype
Start for free. Upgrade anytime.
MonthlyAnnually
Shipped with pre.dev
Real apps planned, built, and deployed end to end.
pre.dev Labs
Built by users
Trusted by 10,000+ founders and engineering teams
Alex Thompson
Founder & CEO at StartupFlow
It planned the whole build before touching code, then ran for hours and opened clean PRs. I reviewed diffs instead of babysitting prompts.
Maria Rodriguez
Staff Engineer at InnovateLab
It handled a multi-file refactor across our codebase without losing context halfway through. The persistent memory is the part other agents get wrong.
David Kim
CTO at TechVenture
Nothing gets marked done until it passes type checks, lint, and browser verification. We stopped getting "looks good but is broken" PRs.
Nicolas Zander
Founder at CleverCompliance
The upfront planning is what sold me. It decomposes the work into a dependency graph before it writes a line. Mindblowingly good.
Max Kryzhanovskiy
Lead Engineer at MosCreative
I've run it across three different model providers depending on the job. Being model-agnostic means I'm never locked in or overpaying on tokens.
Alex Thompson
Founder & CEO at StartupFlow
It planned the whole build before touching code, then ran for hours and opened clean PRs. I reviewed diffs instead of babysitting prompts.
Maria Rodriguez
Staff Engineer at InnovateLab
It handled a multi-file refactor across our codebase without losing context halfway through. The persistent memory is the part other agents get wrong.
David Kim
CTO at TechVenture
Nothing gets marked done until it passes type checks, lint, and browser verification. We stopped getting "looks good but is broken" PRs.
Nicolas Zander
Founder at CleverCompliance
The upfront planning is what sold me. It decomposes the work into a dependency graph before it writes a line. Mindblowingly good.
Max Kryzhanovskiy
Lead Engineer at MosCreative
I've run it across three different model providers depending on the job. Being model-agnostic means I'm never locked in or overpaying on tokens.
Alex Thompson
Founder & CEO at StartupFlow
It planned the whole build before touching code, then ran for hours and opened clean PRs. I reviewed diffs instead of babysitting prompts.
Maria Rodriguez
Staff Engineer at InnovateLab
It handled a multi-file refactor across our codebase without losing context halfway through. The persistent memory is the part other agents get wrong.
David Kim
CTO at TechVenture
Nothing gets marked done until it passes type checks, lint, and browser verification. We stopped getting "looks good but is broken" PRs.
Nicolas Zander
Founder at CleverCompliance
The upfront planning is what sold me. It decomposes the work into a dependency graph before it writes a line. Mindblowingly good.
Max Kryzhanovskiy
Lead Engineer at MosCreative
I've run it across three different model providers depending on the job. Being model-agnostic means I'm never locked in or overpaying on tokens.
Alex Thompson
Founder & CEO at StartupFlow
It planned the whole build before touching code, then ran for hours and opened clean PRs. I reviewed diffs instead of babysitting prompts.
Maria Rodriguez
Staff Engineer at InnovateLab
It handled a multi-file refactor across our codebase without losing context halfway through. The persistent memory is the part other agents get wrong.
David Kim
CTO at TechVenture
Nothing gets marked done until it passes type checks, lint, and browser verification. We stopped getting "looks good but is broken" PRs.
Nicolas Zander
Founder at CleverCompliance
The upfront planning is what sold me. It decomposes the work into a dependency graph before it writes a line. Mindblowingly good.
Max Kryzhanovskiy
Lead Engineer at MosCreative
I've run it across three different model providers depending on the job. Being model-agnostic means I'm never locked in or overpaying on tokens.
Cynthia Colmenares
Founder at Jack Frost Design
Every task lands as its own feature branch and PR—it never pushes to main. It slots into the exact GitHub workflow our team already runs.
Ksenia Snegireva
CTO at Y-Tech
Each agent gets exactly the context it needs in an isolated sandbox—not our whole repo crammed into one window. The token savings on long runs are real.
Gleb Tertychnyi
Founder at WPWP
I pointed it at our existing repo and it reverse-engineered the architecture and kept building in our patterns. No starting from scratch.
Rick Wong
Founder at DataSaa
It kicked off the whole roadmap in one run and the agents worked through the dependency graph on their own. Genuinely long-horizon, not a glorified autocomplete.
Cynthia Colmenares
Founder at Jack Frost Design
Every task lands as its own feature branch and PR—it never pushes to main. It slots into the exact GitHub workflow our team already runs.
Ksenia Snegireva
CTO at Y-Tech
Each agent gets exactly the context it needs in an isolated sandbox—not our whole repo crammed into one window. The token savings on long runs are real.
Gleb Tertychnyi
Founder at WPWP
I pointed it at our existing repo and it reverse-engineered the architecture and kept building in our patterns. No starting from scratch.
Rick Wong
Founder at DataSaa
It kicked off the whole roadmap in one run and the agents worked through the dependency graph on their own. Genuinely long-horizon, not a glorified autocomplete.
Cynthia Colmenares
Founder at Jack Frost Design
Every task lands as its own feature branch and PR—it never pushes to main. It slots into the exact GitHub workflow our team already runs.
Ksenia Snegireva
CTO at Y-Tech
Each agent gets exactly the context it needs in an isolated sandbox—not our whole repo crammed into one window. The token savings on long runs are real.
Gleb Tertychnyi
Founder at WPWP
I pointed it at our existing repo and it reverse-engineered the architecture and kept building in our patterns. No starting from scratch.
Rick Wong
Founder at DataSaa
It kicked off the whole roadmap in one run and the agents worked through the dependency graph on their own. Genuinely long-horizon, not a glorified autocomplete.
Cynthia Colmenares
Founder at Jack Frost Design
Every task lands as its own feature branch and PR—it never pushes to main. It slots into the exact GitHub workflow our team already runs.
Ksenia Snegireva
CTO at Y-Tech
Each agent gets exactly the context it needs in an isolated sandbox—not our whole repo crammed into one window. The token savings on long runs are real.
Gleb Tertychnyi
Founder at WPWP
I pointed it at our existing repo and it reverse-engineered the architecture and kept building in our patterns. No starting from scratch.
Rick Wong
Founder at DataSaa
It kicked off the whole roadmap in one run and the agents worked through the dependency graph on their own. Genuinely long-horizon, not a glorified autocomplete.
Still have questions? We've got answers
Your next build is one prompt away
Hand off the whole build and review the PRs.