FAQ
Short answers to common questions about Cusp, models, pricing, privacy and bringing your existing setup.
General
What is Cusp?
An agentic coding workspace that works with any model. It reads your code, edits files, runs commands and checks its work, using the provider and model you choose. It runs as a desktop app and a CLI on one shared runtime. Crew, AI teammates that work on a schedule, runs locally today; the web app and cloud crews are coming soon.
Is Cusp ready for daily use?
Cusp 0.1 is an early preview. The core runtime, harness, provider layer, CLI and desktop shell exist and are tested, but expect rough edges and changes before 1.0. Commit your work before long sessions, and read the changelog when you update.
What does Cusp cost?
Cusp needs an account on every plan. The Free plan is $0 and includes 2,500 agent actions a month on your own keys, plus $1 a month of efficient Cusp models and a $3 welcome credit. Pro is $20 a month ($16 billed yearly) and includes Cusp models every week (100% of a month is $10 of usage at our cost), with unlimited use of your own keys. Max, Team and Enterprise are on the pricing page.
With your own keys, you pay your model provider directly, and nothing is added to your provider's bill. For Cusp models, usage inside your plan counts at the model provider's published price, with cache discounts passed through; usage beyond your plan is charged at list price × 1.05, which includes a disclosed 5% fee.
Models
Which models work best?
Frontier models from Anthropic, OpenAI and Google do best on long, open-ended tasks. Open-weight models such as GLM, DeepSeek, Kimi, Qwen and MiniMax are strong and much cheaper, and Cusp's harness adapts to them. Cusp picks a strong default when you add a provider; Models and effort shows typical picks.
How does Cusp handle different models?
Cusp works with any endpoint that speaks the OpenAI, Anthropic or Gemini API, and adapts its tools to each model family: apply_patch for GPT models, str_replace-style edits for Claude and Gemini, and hash-anchored line edits for weaker and local models. It replays each provider's reasoning state and repairs malformed tool calls from open models. See Tools and Providers and gateways.
Can I use a local model?
Yes. Point Cusp at Ollama, LM Studio, vLLM or any OpenAI-compatible server. Choose a model that supports tool calling and give it at least 32k tokens of context. Expect smaller models to need more guidance.
Can I use my company's gateway?
Yes, if it speaks the OpenAI, Anthropic or Gemini API. Enter its base URL and key. Gateways that need extra headers (Portkey, Cloudflare) can be configured with them.
Can I switch models in the middle of a session?
Yes, with /model or the model picker. The conversation carries over. The first turn on the new model won't benefit from the previous provider's prompt cache.
Does Cusp work with my Claude Pro or Max subscription?
No. Cusp connects with API keys from providers and gateways. It doesn't sign in with consumer chat subscriptions.
Your existing setup
Does Cusp work with the setup I already have?
Yes. Cusp reads AGENTS.md, CLAUDE.md, GEMINI.md, .claude/settings.json, skills, subagents, commands, hooks and .mcp.json where they are, so there is nothing to migrate, and you can keep using other agents on the same repository. Configuration that Cusp doesn't read directly comes over with the import: in the desktop app's setup, or with cusp import. Plugins in the .claude-plugin format install with cusp plugin (see Hub). Details are in Import your setup.
Privacy and security
Does my code go to Cusp's servers?
With your own keys, no. The daemon sends model requests directly to the provider you chose, and your code and session history stay on your machine unless you use a cloud feature such as sync. With Cusp models, requests pass through Cusp's managed gateway to the provider; we keep usage metadata such as token counts and cost, not the content. You need a Cusp account either way. The Privacy Policy lists exactly what reaches Cusp's servers; see also Security and privacy.
Does Cusp collect telemetry?
Crash reports and usage metrics are off by default. Cusp asks once, and you can change your choice in Settings → Privacy. See Security and privacy.
Can the agent delete my files or run dangerous commands?
Only with your approval, unless you've pre-approved it with rules or a permissive mode. Edits are checkpointed so you can rewind them. Shell commands can be sandboxed. Writes to .git, agent configuration and shell startup files need your approval in every mode. Shell side effects such as deleted directories or pushed commits can't be rewound, so be deliberate about what you allow.
Platforms
Which platforms are supported?
The desktop app runs on macOS 12 and later, Windows 10 and 11, and x86_64 Linux. The CLI runs on macOS, Linux and Windows. The OS sandbox for shell commands is available on macOS and Linux.
Is there an IDE extension?
In preview. The VS Code extension puts sessions, diffs and approvals in a sidebar, and Zed, JetBrains and other editors that speak the Agent Client Protocol can run Cusp through cusp acp. Neither is in a marketplace yet. Cusp also works alongside any editor: file paths in the transcript open in your editor, and the CLI runs in any terminal, including the one inside your IDE.