No token markup
Use your own provider accounts at their own rates. Pay Pumpkin for the editor, with no added charge on direct model tokens.
Explore featureThe Pumpkin feature guide
A whole editor.
A lot more possibility.
From the first question to the finished work. Choose your models, shape your process, and put a full set of tools to work on your machine.
No markup with your own keysDirect model connectionsIndependent of model vendors
Your editor. Your choice.
Keep control of your models, your spend, and where your work goes.
Direct connections, local work
Run the agent on your machine and connect to models with your own keys. Know exactly when optional services send data elsewhere.
Explore featureIndependent of model vendors
Built by developers, for developers. Not owned by Elon Musk, Sam Altman, or Dario Amodei. Choose models on their merits.
Explore featureModels and providers
Choose from OpenAI, Anthropic, xAI, Gemini, Meta, Fireworks, Amazon Bedrock, supported subscriptions, and a local model.
Explore featureUse subscription credits first
Put supported Codex and Claude Code subscription quota ahead of paid API routes, with provider order you control.
Explore featureAuto mode, tuned by you
Let Pumpkin choose a model for the task. Tune eligible models, skill ranges, ratings, and descriptions in Settings.
Explore featureA model on your machine
Run a local coding model and use it for small helper tasks. Choose a cloud stand-in when local hardware is not a good fit.
Explore featureGet to the right context
Find the code, keep the history, and give the agent what it needs.
Semantic code search
Find code by what it does, then trace callers and callees. Keep the index current with local or chosen cloud embeddings.
Explore featureProject memory
Keep project facts, feature notes, and decisions in files the agent can find again across chats.
Explore featureChats and context
Organize chats with projects, pins, search, and archives. Keep long work readable with saved history and context controls.
Explore featureVoice and attachments
Speak a request, attach files and images, and give the agent the context it needs without retyping everything.
Explore featureWork your way
Shape the process, split the work, and decide how far the agent should go.
Plans, goals, and progress
Turn broad requests into reviewable plans, track steps as they finish, and keep a clear goal through long work.
Explore featureYour development process
Define the steps the agent follows, from research and planning to tests, review, shipping, and live checks.
Explore featureSettings that fit you
Tune communication, comments, tests, questions, effort, and the point where the agent should stop.
Explore featureSubagents and side threads
Split complex work across specialist agents, follow their progress, and ask side questions without losing the main task.
Explore featureRules, skills, and commands
Teach repeatable workflows with project rules, reusable skills, custom agents, and quick commands.
Explore featureHooks, permissions, and trust
Add checks around agent actions, inspect permission requests, and keep untrusted workspaces restricted.
Explore featureAutopilot
Let a model handle follow-up choices in a chat, with visible actions, a round limit, and a Stop control.
Explore featureIsolated worktrees
Give each task its own branch and checkout so parallel work has space to proceed.
Explore featureCheckpoints and rewind
Return workspace files to an earlier turn when you change a request, with Git-backed checkpoints.
Explore featureMore than a chat box
Work with real browsers, live visuals, connected tools, and your full editor.
A browser the agent can use
Open real browser pages, inspect behavior, capture screenshots and recordings, and take over when you need to.
Explore featureLive canvases
Turn an answer into an interactive chart, dashboard, comparison, or small tool beside your chat.
Explore featureConnected tools with MCP
Connect the agent to the services and tools you use, with visible setup, sign-in state, and enable controls.
Explore featureWeb research
Search the web, read sources, and connect fresh evidence to the task you are working on.
Explore featureA full desktop editor
Use files, terminals, Git, debugging, extensions, and AI work together on Windows, macOS, and Linux.
Explore featureBuild and test coordination
Run focused checks, keep long jobs visible, and share machine resources across concurrent agent work.
Explore featureKeep work within reach
Move between machines, check in from your phone, and keep recurring work moving.
Remote access
Check chats and guide work from your phone through approved, encrypted device connections.
Explore featureMultiple computer sync
Sync chats, attachments, plans, and eligible settings through storage you control. Keep secrets and machine paths separate.
Explore featureWSL, SSH, and paired machines
Work with projects on Linux environments and remote hosts while keeping the desktop editor close at hand.
Explore featureLoops
Run recurring agent work on a schedule, with saved memory, visible runs, controls, and optional Slack updates.
Explore featureFollow-ups and smart waits
Schedule a later task or let a chat resume when a job, file, endpoint, or CI run reaches the state it needs.
Explore featureSlack alongside your work
Start work from Slack, receive loop summaries, and answer questions while the agent runs on your linked machine.
Explore featureSee what is happening
Track actions, tokens, spend, and team use without guesswork.
Usage beside the work
See tokens, estimated cost, model routes, and cache use alongside the chat that produced them.
Explore featureDetailed usage reports
Review usage over time, filter by user and model, inspect charts, and export data from your chosen log sources.
Explore featureTeam seats and shared standards
Manage business seats, invite teammates, and share project guidance and usage policy while keeping individual accounts.
Explore featureMake it your own.
One editor for Windows, macOS, and Linux. Start with the tools you need, and grow from there.
Get Pumpkin See pricing