pulsebot 0.3.0

A lightweight, stream-native AI agent powered by Timeplus

PulseBot

Stream-native AI Agent powered by Timeplus

PulseBot is a lightweight, extensible AI agent framework that uses Timeplus streaming database as its backbone for real-time message routing, observability, and memory storage.

✨ Features

• Stream-Native Architecture - All communication flows through Timeplus streams
• Multi-Provider LLM Support - Anthropic Claude, OpenAI, OpenRouter, Ollama, and NVIDIA
• Vector Memory - Semantic search using embeddings stored in Timeplus
• SQL-Native Scheduling - Timeplus Tasks replace traditional cron jobs
• Interactive Workspaces - Build and publish dynamic artifacts and runnable web apps
• Extensible Skills - Plugin-based tool system (web search, file ops, shell, workspace)
• Multi-Channel - Telegram, webchat, with easy extension to Slack/WhatsApp
• Real-Time Observability - All LLM calls and tool executions logged to streams
• Production Ready - Docker deployment, async architecture, structured logging

🏗️ Architecture

🚀 Quick Start

Prerequisites

• Python 3.11+
• Timeplus or Proton

Installation

# Clone repository
git clone https://github.com/timeplus-io/PulseBot.git
cd PulseBot

# Install package
pip install -e .

# Generate default config
pulsebot init

Configuration

Edit config.yaml with your API keys:

agent:
  name: "PulseBot"
  model: "claude-sonnet-4-20250514"  # or use ollama model

providers:
  anthropic:
    api_key: ${ANTHROPIC_API_KEY}
  
  # For local testing with Ollama
  ollama:
    enabled: true
    host: "http://localhost:11434"
    default_model: "llama3"

workspace:
  base_dir: "./workspaces"
  workspace_port: 8001
  internal_api_key: "${WORKSPACE_INTERNAL_KEY}"

Using Ollama (Local Testing)

# Install Ollama (macOS)
brew install ollama

# Pull a model
ollama pull llama3

# Start Ollama server
ollama serve

# Update config.yaml to use Ollama
# Set agent.provider: "ollama" and agent.model: "llama3"

Setup & Run

# Start the agent (streams are initialized automatically on first run)
pulsebot run

# Or start the API server
pulsebot serve

🐳 Docker Deployment

# Set environment variables
export ANTHROPIC_API_KEY=your_key
export TELEGRAM_BOT_TOKEN=your_token

# Start all services
docker-compose up -d

This starts:

• Timeplus - Streaming database (ports 8123, 3218, 8463)
• PulseBot Agent - Message processing
• PulseBot API - REST/WebSocket interface (port 8000)

📖 CLI Commands

Command	Description
pulsebot run	Start the agent loop
pulsebot serve	Start FastAPI server
pulsebot chat	Interactive CLI chat
pulsebot init	Generate config.yaml
pulsebot task list	List scheduled tasks

🔧 Built-in Skills

Skill	Tools	Description
web_search	web_search	Brave Search / SearXNG integration
file_ops	read_file, write_file, list_directory	Sandboxed file operations
shell	run_command	Shell execution with security guards
workspace	workspace_create_app, workspace_write_file, ...	Create and publish dynamic artifacts and web apps

AgentSkills.io Support

PulseBot supports the agentskills.io standard for external skill packages. Skills are discovered from configured directories by scanning for SKILL.md files.

Configure skill directories in config.yaml:

skills:
  skill_dirs:
    - "./skills"
    - "/shared/skills"
  disabled_skills: []

Create a skill package as a directory with a SKILL.md file:

skills/
  my-skill/
    SKILL.md          # Required: YAML frontmatter + instructions
    scripts/          # Optional: executable code
    references/       # Optional: supplementary docs

The SKILL.md uses YAML frontmatter for metadata and Markdown body for instructions:

---
name: my-skill
description: Does something useful when the user asks about X.
---

# My Skill

Full instructions loaded on demand by the agent.

Only skill name and description are loaded into the system prompt at startup (~24 tokens per skill). Full instructions are loaded on demand when the agent calls the load_skill tool.

Adding Custom Code Skills

from pulsebot.skills import BaseSkill, ToolDefinition, ToolResult

class MySkill(BaseSkill):
    name = "my_skill"

    def get_tools(self) -> list[ToolDefinition]:
        return [ToolDefinition(
            name="my_tool",
            description="Does something useful",
            parameters={"type": "object", "properties": {}}
        )]

    async def execute(self, tool_name: str, args: dict) -> ToolResult:
        return ToolResult.ok("Success!")

📡 API Endpoints

Web Chat UI

Access the built-in web chat interface at http://localhost:8000/ after starting the API server.

REST & WebSocket Endpoints

Method	Endpoint	Description
GET	/	Web chat UI
GET	/health	Health check
POST	/chat	Send message (async)
WS	/ws/{session_id}	Real-time chat
GET	/sessions/{id}/history	Get conversation history

🗄️ Timeplus Streams

Stream	Purpose
messages	All agent communication (user input, agent response, tool calls)
llm_logs	LLM call observability (tokens, latency, cost)
tool_logs	Tool execution logging (name, arguments, duration, status)
memory	Vector-indexed memories with semantic search
events	System events & alerts

🔐 Environment Variables

# Required (one LLM provider)
ANTHROPIC_API_KEY=...     # For Claude models
# or
OPENAI_API_KEY=...        # For OpenAI models

# Timeplus
TIMEPLUS_HOST=localhost
TIMEPLUS_PASSWORD=...

# Optional
OPENAI_API_KEY=...        # Also used for memory embeddings
TELEGRAM_BOT_TOKEN=...    # For Telegram channel

📚 Documentation

• Technical Design - Full architecture documentation
• Configuration Guide - All settings and environment variables
• Agent Workspace - Dynamic artifacts and full-stack apps
• Telegram Setup - Connect PulseBot to Telegram
• Memory System - Vector memory and embeddings
• Skills System - Plugin architecture