BustaAI: one chat for Cursor, Codex and OpenAI agents

Three modes, one conversation

The panel greets you with the same choice every time: one conversation, three ways to work. These are the in-app descriptions of each mode.

• Ask

  Ask about the project. It can read and research, but it doesn’t modify files.

• Plan

  Analyses the project and writes a plan in Markdown before touching anything.

• Agent

  Reads, edits, runs commands and iterates within your permissions.

Cursor, Codex and the OpenAI API

BustaAI gives supported providers one interface, but it doesn’t replace their engines: each provider uses its own execution engine.

• Cursor

  Integrated through Cursor’s official CLI and ACP.

• Codex

  Integrated through the Codex app-server.

• OpenAI API

  Direct OpenAI API integration.

Model availability, account requirements and capabilities depend on the provider. Cursor and Codex integrations rely on their respective official tools and authentication.

The composer

• Describe a task and type @ to attach a file, an error, a terminal, a container or an MCP.
• Pick the mode (Ask, Plan or Agent) and the model right from the composer.
• The active provider and model are also shown in the status bar — for example “Codex · GPT-6.1-Sol” in the screenshot.

Project memory

Structured, editable context with what the agents need to know about the project.

• Project goals
• Decisions
• Instructions
• Relevant files
• Git state

Parallel conversations

Run multiple agent chats in tabs, each with its own status indicator and conversation history. The sidebar lets you start a new chat, search past chats and browse them by folder.

Privacy and approvals

• Project context and BustaAI’s local state are managed on your device.
• Relevant context is sent to the selected AI provider when you make a request; your provider’s policies also apply.
• Tool execution and permissions depend on the selected agent engine. BustaAI presents approval requests and activity information where supported.