FAQ

General

What makes ZeroSphere different?

Every other AI IDE stops at the file. ZeroSphere compiles your code, launches the application, clicks through your UI, catches what broke, and fixes it. The agent sees your running app the same way you do — raw pixels, pure vision. No other tool does this.

Is ZeroSphere open source?

No. ZeroSphere is proprietary software. The source code is not publicly available.

Which models work?

Any model with an OpenAI-compatible endpoint. Anthropic Claude, OpenAI GPT, Google Gemini, or any compatible provider via OpenRouter or a custom profile.

Do I need my own API key?

Yes. ZeroSphere is BYOK (Bring Your Own Key). Your keys are stored locally and requests go directly from your machine to your provider. ZeroSphere is not in that loop.

What does it cost?

ZeroSphere itself is free during early access. You pay your model provider directly at their standard rate.

Capabilities

Can the agent modify files?

Yes, with your permission. Create, edit, and delete are all controlled in Settings > AI.

Can the agent run terminal commands?

Yes, with your permission. Builds, test suites, dev servers, and shell commands are controlled via run_command_mode in config.toml.

Can the agent see my screen?

No. It sees only the virtual display sessions it creates. Your actual screen is never visible to the agent — unless you explicitly grant a main display session, in which case you see a STOP button the entire time.

What languages does it support?

Any language. Language support comes from the model, not ZeroSphere.

Can it search the web?

Yes, when TAVILY_API_KEY is configured. The agent uses web search to look up docs, error messages, and changelogs.

Context and Memory

Does the agent remember previous conversations?

No. Each session starts fresh. The agent sees the current conversation, any files you reference, and your codebase. Nothing carries over from previous sessions unless memory is saved.

Why did it forget something I told it?

Very long conversations hit context limits and older messages drop off. Start a new session for unrelated tasks. Put persistent instructions in AGENTS.md in your project root so they are always loaded.

How do I give it more context?

Attach files in the conversation, reference file paths explicitly, or add project-wide instructions to AGENTS.md.

Permissions and Security

Is my code sent to AI providers?

Yes, to your chosen provider directly from your machine. It does not go through ZeroSphere servers. Subject to your provider's privacy policy.

Can I prevent it from accessing certain files?

Add them to .gitignore or .agentignore in your project root. Set file access to ask_each_time and deny access when prompted.

Recommended permission setup:

[permissions.terminal]
run_command_mode = "let_model_decide"
os_run_in_session_mode = "ask_each_time"

[permissions.os_tools]
os_create_virtual_session_mode = "ask_each_time"
os_create_main_display_session_mode = "ask_each_time"
os_bind_window_mode = "let_model_decide"

Getting Better Results

The agent keeps making mistakes: Attach the relevant files. Be specific about what you want. Break large tasks into smaller ones.

It is not following my instructions: Add persistent rules to AGENTS.md. Rephrase to be more explicit. Start a new session if the conversation has grown very long.

Responses are too verbose: Tell it directly: "Make the change, no explanation." Or add "Be concise" to AGENTS.md.

It broke something: Revert with git. Tell the agent what broke and why. Try again with more context.

Advanced

Can I use multiple models in one session?

No. One model per session. Start separate sessions to compare models.

Can the agent learn from my corrections?

No. Each session is independent. Add corrections and patterns to AGENTS.md so the agent sees them every session.

Can I extend the agent's capabilities?

Yes, via MCP servers. See MCP Configuration for setup.