Initialize Economic Tracker Args: signature: Agent signature/name initial_balance: Starting balance in dolla

Initialize wrap-up workflow Args: llm: Language model for decision-making (if None, creates default) logger:

Initialize LiveAgent Args: signature: Agent signature/name basemodel: Base model name initial_ba

Initialize MCP client Args: config: Dictionary mapping server names to their configuration Each confi

Build the evaluation prompt for the LLM

ClawMode — nanobot gateway with ClawWork economic tracking.

Use LLM to decide which artifacts to submit

Download chosen artifacts from sandbox to local storage

从 PDF 中提取指定页面并返回单页 PDF 的字节数据。 page_num 从 0 开始。

List all artifacts in the active sandbox.

将 PDF 全部页面转换为 PNG base64 列表，用于 Qwen image_url。 依赖 pdf2image 和 poppler；若未安装则给出明确错误。 逐页转换，确保多页文件不会被截断。

Read artifact file contents with size limits and format extraction Args: artifact_paths: List of artifact file paths

Register all nanobot tools plus ClawWork tools.

Determine if we should download artifacts

Submit downloaded artifacts for evaluation

Add profit/loss from trading Args: profit: Profit amount (negative for loss) description: Optional descripti

Chat with the agent locally, with ClawWork economic tracking. Works like ``nanobot agent`` but every message is cost-tracked and the /clawwor

Endpoint for LiveBench to broadcast updates to connected clients This should be called by the LiveAgent during execution

Clean up provider resources.

Create a file with the given content. Useful for creating work artifacts. Args: file_path: Path where the file should be created

Create a file in the sandboxed work directory. Supports multiple formats: txt, md, csv, json, xlsx, docx, pdf Args: filename: N

Create a video from text slides and/or images. Args: slides_json: JSON string describing slides. Format: [

Decide your daily activity: work or learn. Args: activity: Must be "work" or "learn" reasoning: Explanation for your decisio

Decide your daily activity: work or learn. Args: activity: Must be "work" or "learn" reasoning: Explanation for your decisio

Execute code in a persistent sandbox with artifact download support. Features: - Provider selection via CODE_SANDBOX_PROVIDER (e2b defau

Execute code in a sandboxed environment with safety restrictions. SECURITY FEATURES: - Execution timeout (30 seconds) - Restricted t

Format cost update message to inject into conversation Args: session_cost: Cost of current session/interaction daily_cost: T

Start the nanobot gateway with ClawWork economic tracking. All configuration is read from ~/.nanobot/config.json. Enable ClawWork by setting

Get detailed information about a specific agent

Get economic metrics for an agent

Get agent's learning memory

Get all tasks assigned to an agent. Uses task_completions.jsonl as the authoritative list of tasks (no duplicates). task_details are looked u

Get list of all agents with their current status

Serve an artifact file for preview/download

Get display name mapping {signature: display_name}

Get current economic status including balance, token costs, and survival status. Returns: Dictionary with economic metrics:

Get list of hidden agent signatures

Get leaderboard data for all agents with summary metrics and balance histories

Generate learning-specific prompt Args: date: Current date signature: Agent signature Returns: Learning prompt

Read the agent's memory file containing previously learned information. Returns: The content of the agent's memory file

Get a random sample of agent-produced artifact files

Get recent error entries

Get recent warning entries

Get your current economic status and balance. Returns: Dictionary with current status information

Get task details by task ID Args: task_id: Task identifier Returns: Task dictionary or None if not

Get full details of today's work task including prompt and reference files. Returns: Complete task information with prompt, referenc

Get the task prompt text Args: task: Task dictionary Returns: Task prompt string

Get statistics about the task dataset Returns: Dictionary with statistics

Get a brief summary of the task Args: task: Task dictionary Returns: Summary string

Get terminal log for an agent on a specific date

Get all tools from all configured MCP servers

DEPRECATED: Trading functionality has been disabled. LiveBench now focuses on work and learn capabilities only. This function is kept fo

Get history of completed work tasks and earnings. Returns: List of past work evaluations and total earnings

Generate detailed work task prompt Args: task: Task dictionary from gdpval reference_files: List of reference file paths

Learn something new and add it to your knowledge base. Args: topic: Topic or title of what you learned knowledge: Detailed k

Use web search to learn about any topic. Searches the web for information and optionally saves the learned information to the agent's memory.

Convenience function to log debug

Convenience function to log info

Convenience function to log warning