Discover AI models, system tools, and community skills.
Google’s next-gen lightweight multimodal model.
Balanced performance and speed for the 2.5 generation.
Advanced reasoning and coding capabilities.
Maximum intelligence for complex tasks.
The latest evolution in the GPT series.
Neuralhub ultimate stateful default environment (Node + Python + VNC Computer Use).
Autonomous AI Engineer (OpenCode CLI)
Google Gemini development environment
Standard TypeScript/Python persistent environment
TypeScript-based AI agent engineering platform with Memory, RAG, and MCP
Secure, sandboxed in-memory Bash environment for AI agents by Vercel Labs
Scrape a URL and return its content in Markdown format.
Start a crawler for a website. Returns a Job ID to track.
Write text content to a file in the workspace storage. Automatically registers as a task artifact. Overwrites if exists.
Read the text content of a file from workspace storage.
List files in a directory.
Semantic search for file content across the workspace. Uses vector embeddings to find relevant code or text.
Mark the task as completed and provide the final answer/artifacts. Always use this to end the conversation.
Pause execution to ask the user for clarification or confirmation.
Delegate a complex task to a specialized sub-agent. This spawns a new agent process in a sub-folder.
Ask a specific question to another agent or expert system (simulated). lightweight version of delegate.
Retrieve a value from the workspace short-term memory (Key-Value).
Store a value in short-term memory. Persists for the session.
Store a fact or text snippet into long-term semantic memory.
Search long-term memory for relevant facts by semantic similarity.
Add a new node (task) to the execution graph. Can be linked to a parent or dependencies.
Get dependencies for a specific task (upstream blockers) or children (downstream).
Add a blocking dependency between two existing tasks (Parent blocks Child).
Write a record to a specific named database (which is just a folder).
Read a record from a named database.
List keys in a named database.
Lists all available resources (agents, databases, compute) for the current workspace.
Reads the recent stream of a specific resource to see what it has been doing.
Posts a message to a specific resource thread to delegate a task.
Fetches the state JSON of a specific running resource.
Executes a shell command inside the Daytona sandbox attached to the current context. All commands should be non-blocking or complete within 10 seconds. You must have a sandboxId in context to use this.
Reads the content of a file inside the Daytona sandbox attached to the current context.
Lists the safe, filtered files in a directory inside the Daytona sandbox (ignores node_modules, .git, etc.).
Writes content to a file inside the Daytona sandbox.
Pauses or hibernates the specific Daytona sandbox to save compute resources. Use this when you are completely finished with your tasks for the user.