Here are some examples:
- Meeting-prep assistant: https://www.youtube.com/watch?v=KZTP4xZM2DY
- Customer support assistant: https://www.youtube.com/watch?v=Xfo-OfgOl8w
- Gmail and Reddit assistant: https://www.youtube.com/watch?v=6r7P4Vlcn2g
Rowboat is open-source (https://github.com/rowboatlabs/rowboat) and has a growing community. We first launched it on Show HN a few months ago (https://news.ycombinator.com/item?id=43763967).
Today we are launching a major update along with a cloud offering. We’ve added built-in tool integrations for 100s of tools like Gmail, Github and Slack, RAG with documents and URLs, and triggers to invoke your assistant based on external events.
Our cloud version includes all the features of the open-source IDE, but runs instantly with no setup or API keys. For launch, we're offering $10 free usage with Gemini models so you can start building right away for free without adding any card details. Paid plans start at $20/month and give you access to additional models (OpenAI, Anthropic, Gemini, with more coming) and higher usage limits.
There’s a growing view that some tasks are better handled by single agents (https://news.ycombinator.com/item?id=45096962), while others benefit from multi-agent systems for higher accuracy ( https://www.anthropic.com/engineering/multi-agent-research-s...). The difference often comes down to scope: a focused task like coding suits a single agent, but juggling multiple domains such as email, Slack, and LinkedIn is better split across agents. Multi-agent systems also help avoid context pollution, since LLMs lose focus when asked to handle unrelated tasks. In addition, cleanly dividing responsibilities makes each agent easier to test, debug, and improve.
However, splitting work into multiple agents and getting their prompts right is challenging. OpenAI and others have published patterns that work well for different scenarios (https://cdn.openai.com/business-guides-and-resources/a-pract...). We’ve added agent abstractions, built on top of OpenAI’s Agents SDK, to support these patterns. These include user-facing agents that can decide to hand off to another agent when needed; task agents that perform internal tasks; and pipelines that deterministically call a sequence of agents.
Rowboat’s copilot (‘Skipper’) is aware of these patterns and has been seeded with tested patterns, such as a manager‑worker setup for a customer support bot, a pipeline for automated document summarization, and multi‑agent workflows for combining web search with RAG. It can:
- Build multi-agent systems from a high-level request and decide how work must be delegated across agents
- Edit agent instructions to make correct tool calls using Composio tools or any connected MCP server
- Observe your playground chat and improve agents based on your tests
We see agentic systems as a spectrum. On one end are deterministic workflows with a few LLM calls. On the other end are fully agentic systems where the LLM makes all control flow decisions - we focus on this end of the spectrum, while still allowing deterministic control where necessary for real-world assistant use cases. We intentionally avoided flowchart-style editors (like n8n) because they become unwieldy when building and maintaining highly agentic systems.
We look forward to hearing your thoughts!
nextworddev•4mo ago
akhisud•4mo ago
Rowboat is especially designed for agentic patterns (e.g. manager-worker) which lend more autonomy to agents. Rowboat's copilot is empowered to organize and orchestrate agents flexibly, based on the nature of the assistant.
KaoruAoiShiho•4mo ago
segmenta•4mo ago
Here is some personal experience: we previously built Coinbase's automated chatbot and we used a flowchart type builder to do that. This was a intent-entity based system that used deep learning models. It started great, but pretty quickly it became a nightmare to manage. To account for the fact that users could ask things out of turn or move across topics every other turn - we added in concepts called jumps - where control could go from one path to another unrelated path of workflow in on hop - which again introduced a lot of maintenance complexity.
The way we see it is that, when we assign a task to another human or a teammate we don't give them a flowchart - we just give them high level instructions. Maybe that should be the standard for building systems with LLMs?
Is this making sense?
zoomzoom•4mo ago
segmenta•4mo ago
No, the instructions are not compiled into a flowchart under the hood. We use OpenAI’s agent SDK and use handoffs as a mechanism to transfer control between agents.
There are 3 types of agents in Rowboat: 1. Conversational agents are ones which can talk to the user. They can call tools and can choose to handoff control to another agent if needed. 2. Task agents can’t talk to users but can otherwise call tools and do things in a loop - they are internal agents. 3. Pipeline agent is a sequence of task agents (here the transfer of control is deterministic).
For instance, if we build a system for airline customer support, there might be a set of conversational agents each for different high level topics like ticketing, baggage etc. and internally they can use task and pipeline agents as needed.
Does this make sense?