Use your plan first
Put subscription quota ahead of paid API use. Requests that would add to a per-token bill draw on your plan instead.
Desktop AI editor for Windows, macOS, and Linux
Your editor. Your models. No token markup.
$20 a month for the editor. Bring your supported Codex or Claude Code subscription, or use your own keys at provider rates. Connect from your workspace in an editor independent of the model vendors.
Built by developers, for developers. Built on Code-OSS.
- 01No upcharge on tokensYour own keys. Your provider's rates.
- 02Direct model connectionsYour workspace runs the agent and connects to your chosen provider.
- 03Independent of model vendorsMade by Ezoic. Not owned by Elon Musk, Sam Altman, or Dario Amodei.
Your money
No upcharge on tokens
Your seat pays for the editor. Bring your own API keys and the provider bills your account at its own rates. Pumpkin adds no token markup and imposes no Pumpkin request quota.
Already pay for a supported Codex or Claude Code subscription? Put its available quota first, then use a configured API route when needed. You control the provider order.
Optional Pumpkin credits use a separate hosted route with service fees. Provider limits and charges still apply.
How model billing works
Lower token spend
Your subscriptions. A full editor. A smaller token bill.
Already pay for Codex or Claude Code? Bring that account to Pumpkin and put its quota first, so work that would be billed per token runs on the plan you already pay for. Auto can favor lower-cost suitable models, and context keeps task facts and recent work without repeatedly sending raw history. Ezoic reported about $450,000 in July token spend and projects under $25,000 a month with Pumpkin. Read the Ezoic case study.
Keep your workflow
Use subscription models with Pumpkin’s agent tools, custom process, and subagents inside your editor.
Choose what comes next
Auto favors lower-cost suitable models. Add an API route if you want a paid option when quota runs out. You choose the provider order.
Your Pumpkin seat is separate. Your provider’s plan terms and limits still apply; API use is billed separately.
See how to use your subscriptionsYour workspace
Direct connections. Local work.
The agent, file edits, terminal, and local search run on your workspace machine. Your own provider keys connect from that host to the model provider, without a Pumpkin-hosted inference relay. Keys use protected secret storage.
Cloud models receive the prompts and context you send. Optional credits, Remote, sync, Slack, reporting, and connected tools have their own data paths. Choose the services you use.
See where your work goesYour workspaceAgent, files, terminal, local search
Your key →
Your model providerCloud inference at provider rates
Your choice
Independent of model vendors
Built by developers, for developers. Pumpkin is not owned by Elon Musk, Sam Altman, or Dario Amodei. Choose models for the work they do, with no model company owning your editor.
Bring your own models
Use supported OpenAI, Anthropic, xAI, Gemini, Meta, Fireworks, Amazon Bedrock, subscription, and local routes. Pick a model yourself or tune Auto to choose from your preferred models.
Explore models and providers
From first question to finished work
The full agent toolkit
A complete guide, with practical examples, setup, and limits for each feature.
Find code by meaning
Semantic search, call tracing, project memory, and context from earlier chats.
Work your way
Custom steps, rules, skills, comments, tests, and a voice that fits you.
Split the work
Subagents, side threads, worktrees, plans, checkpoints, and clear progress.
Use real tools
A browser, live canvas, MCP connections, web research, and build checks.
Stay in reach
Remote access, multiple computer sync, WSL, SSH, loops, and Slack.
See the cost
Inline tokens and spend, model routes, detailed reports, and team usage.
Why switch?
Keep your editor working for you.
- Use your existing plansBring supported Codex and Claude Code subscriptions. Put their quota first and spend less on tokens.
- Provider-rate tokensNo markup on your own-key model use.
- Local controlYour workspace runs the agent. You choose its connections.
- Independent choiceYour preferred models, process, and tools in one editor.
Windows, macOS, and Linux
Get a licensed account, download Pumpkin, and connect your models.