OnsetLab provides locally runnable AI agents that can plan and execute sequences of tool calls, automatically falling back to step‑by‑step ReAct reasoning when a call fails. The platform supports any Ollama‑served LLM (e.g., Qwen, Mistral, Hermes, Gemma) and offers plug‑and‑play integration with MCP servers for services like GitHub, Slack, Notion, file systems, and web search, all via a simple Python API.
Funding
Funding not disclosed
Founders
Product
Problem
Developers and enterprises often need AI agents that can interact with external tools (e.g., calculators, date-time services, or custom APIs) but must rely on cloud-based services, which introduce latency, privacy concerns, and dependency on external APIs.
Solution
OnsetLab offers locally runnable AI agents that can invoke a configurable set of external tools directly from the user's environment. The agents generate an execution plan, call the required tools in sequence, and automatically switch to a ReAct-style step‑by‑step reasoning mode if a tool call fails. Built‑in self‑correction mechanisms detect and fix common errors such as wrong time zones or malformed inputs, retrying the operation without user intervention. The platform supports a wide range of LLM backends—including Qwen, Mistral, Hermes, Gemma, and any model served via Ollama—allowing users to run the agents on their own hardware without API keys or cloud dependencies. Integration with MCP servers enables seamless access to services like GitHub, Slack, Notion, file systems, and web search through a single configuration line.
Target Audience
OnsetLab is aimed at software developers, data scientists, and IT teams that require privacy‑preserving, offline AI automation for tool integration and workflow orchestration.
Features
- Execution planning that sequences multiple tool calls before runtime
- Automatic fallback to ReAct reasoning when a tool call fails, preserving task continuity
- Self‑correcting loop that detects parameter errors (e.g., timezone mismatches) and retries with adjusted inputs
- Plug‑and‑play support for MCP servers to connect to GitHub, Slack, Notion, filesystem, and web search services
- Compatibility with any Ollama‑served LLM, including Qwen, Mistral, Hermes, and Gemma, enabling on‑premise inference
- Simple Python API (`Agent(model="...", tools=[...])`) and one‑line installation via pip