rowboat

Automatically ingests emails, meeting notes, calendar events, and documents from integrated sources (Gmail, Google Calendar, Fireflies, Granola) and builds a queryable knowledge graph stored as plain Markdown files in an Obsidian-compatible vault (~/.rowboat/). Uses entity extraction and relationship mapping to create interconnected nodes representing people, projects, and topics, enabling semantic search and context retrieval without cloud dependency.

Runs persistent background agents that continuously sync data from external services (Gmail, Google Calendar, Fireflies, Granola) on configurable schedules, transforming heterogeneous data formats into unified Markdown representations. Implements OAuth-based authentication and handles incremental updates to avoid re-processing entire datasets, with error handling and retry logic for failed syncs.

Stores all workflow definitions, agent configurations, prompts, and project settings as Markdown files in the local vault, enabling version control, human readability, and portability. Supports import/export of workflows for sharing and migration, with Markdown as the canonical format for all configuration rather than proprietary binary formats.

Supports multiple isolated projects within a single Rowboat Web Application instance, with separate workflows, configurations, and data for each project. Implements workspace-level access control and configuration, enabling teams to organize agent workflows by project or department without cross-contamination of data or configurations.

Integrates with Twilio to enable voice-based interaction with agents through phone calls or voice messages. Converts voice input to text, processes through agent workflows, and returns voice responses, enabling hands-free agent access for mobile or voice-first use cases.

Provides a Python SDK for building agent workflows programmatically, enabling developers to define agents, tools, and workflows in Python code rather than through UI or configuration files. Supports agent instantiation, tool registration, workflow execution, and result handling through Python APIs.

Implements Rowboat X as an Electron application with inter-process communication (IPC) between main process and renderer process, enabling local-first knowledge graph management and copilot chat on desktop. Uses Electron's native file system access to manage Markdown vault and background agents without cloud dependency.

Provides an interactive chat interface (Skipper backend in Web Application, Copilot Chat in Desktop Application) that uses the local knowledge graph as context to assist with work tasks like meeting prep, email drafting, and document creation. Implements RAG (Retrieval-Augmented Generation) to inject relevant knowledge graph nodes into LLM prompts, enabling responses grounded in user's work history and relationships.

Provides a drag-and-drop UI for designing agent workflows in the Web Application, allowing users to define agent nodes, tool connections, and data flow between agents without code. Stores workflow definitions as structured configuration (likely JSON or YAML) that can be executed by the agent runtime, with support for conditional branching, loops, and parallel execution patterns.

Provides an interactive testing environment in the Web Application where users can execute workflows and observe agent behavior in real-time using Server-Sent Events (SSE) for streaming responses. Enables step-by-step execution inspection, intermediate output viewing, and error diagnosis without deploying to production, with support for multiple test runs and result comparison.

Implements Retrieval-Augmented Generation using Qdrant as the vector store backend (configured in docker-compose.yml), enabling semantic search over knowledge graph and documents. Converts Markdown nodes and ingested documents into embeddings, stores them in Qdrant, and retrieves relevant context for LLM prompts based on semantic similarity rather than keyword matching.

Exposes REST endpoints for triggering agent workflows, querying results, and receiving webhook callbacks from external systems. Enables integration with third-party tools and automation platforms, allowing workflows to be triggered by external events (e.g., Slack messages, form submissions) and results to be pushed to downstream systems without manual intervention.

Supports Model Context Protocol (MCP) servers as a standardized way to connect external tools and data sources to agents. Agents can discover and invoke MCP-compatible tools through a unified interface, enabling integration with any tool that implements the MCP specification without custom adapter code.

Integrates with Composio to provide agents with access to pre-built actions for common SaaS tools (email, calendar, CRM, etc.). Agents can invoke Composio actions without custom tool implementation, reducing development time for common workflows like sending emails, creating calendar events, or updating CRM records.

Provides a command-line interface (rowboatx) for running agents in automated environments without GUI, enabling integration with CI/CD pipelines, cron jobs, and server-side automation. Supports workflow execution, model configuration, and result output in formats suitable for scripting and automation.