OnsetLab

OnsetLab

Tool-calling AI agents that run locally.

[](https://opensource.org/licenses/Apache-2.0) [](https://python.org) [](https://pypi.org/project/onsetlab/) [](https://ollama.com)

[Quick Start](#quick-start) · [Architecture](#architecture) · [MCP Servers](#mcp-servers) · [CLI](#cli) · [Models](#tested-models) · Docs

---

Local models are fast, free, and private. But ask one to call a tool and it falls apart. Wrong function names, broken parameters, infinite loops.

The models are capable. The framework wasn't.

OnsetLab makes 3B-7B models do reliable tool calling through a hybrid REWOO/ReAct architecture. The framework handles planning, execution, and error recovery. The model only does what it's good at: one step at a time.

---

https://github.com/user-attachments/assets/46a99c79-de91-4c29-b8f4-06a43f85957a

Quick Start

pip install onsetlab

Requires Ollama running locally with a model pulled:

ollama pull phi3.5

from onsetlab import Agent
from onsetlab.tools import Calculator, DateTime

agent = Agent("phi3.5", tools=[Calculator(), DateTime()])

result = agent.run("What's 15% tip on $84.50?")
print(result.answer)

The agent routes the query, builds an execution plan, calls the right tool, and returns the answer. No prompt engineering required.

---

Architecture

flowchart TD
    Q["Query"] --> R["Router"]

    R -->|"tools needed"| P["Planner"]
    R -->|"no tools"| D["Direct Answer"]

    P --> E["Executor"]
    E --> S["Solver"]

    P -. "plan fails" .-> RE["ReAct Fallback"]

    D --> A["Answer"]
    S --> A
    RE --> A

    style Q fill:#4a6670,stroke:#4a6670,color:#fff
    style R fill:#fff,stroke:#4a6670,color:#2d3b40
    style P fill:#e8f0fe,stroke:#7aa2f7,color:#3b5998
    style E fill:#e8f0fe,stroke:#7aa2f7,color:#3b5998
    style S fill:#e8f0fe,stroke:#7aa2f7,color:#3b5998
    style D fill:#edf7ef,stroke:#9ece6a,color:#2d6a2e
    style RE fill:#fdf4e7,stroke:#e0af68,color:#8a6914
    style A fill:#4a6670,stroke:#4a6670,color:#fff

The Router classifies queries as tool-needed or direct-answer using the model itself. The Planner generates structured THINK -> PLAN steps with auto-generated tool rules from JSON schemas. The Executor resolves dependencies and runs tools in order. If planning fails, the ReAct Fallback switches to iterative Thought -> Action -> Observation loops to recover.

---

Built-in Tools

| Tool | Description | |------|-------------| | Calculator | Math expressions, percentages, sqrt/sin/log | | DateTime | Current time, timezones, date math, day of week | | UnitConverter | Length, weight, temperature, volume, speed, data | | TextProcessor | Word count, find/replace, case transforms, pattern extraction | | RandomGenerator | Random numbers, UUIDs, passwords, dice rolls, coin flips |

> More tools will be added over time.

MCP Servers

Connect any MCP-compatible server to give your agent access to external tools like GitHub, Slack, Notion, and more.

from onsetlab import Agent, MCPServer

server = MCPServer.from_registry("filesystem", extra_args=["/path/to/dir"])

agent = Agent("phi3.5")
agent.add_mcp_server(server)

result = agent.run("List all Python files in the directory")
print(result.answer)

agent.disconnect_mcp_servers()

Any MCP server available via npm works too. See the docs for examples.

Built-in registry: filesystem · github · slack · notion · google_calendar · tavily

---

CLI

python -m onsetlab                                          # interactive chat
python -m onsetlab --model qwen2.5:7b                       # specify model
python -m onsetlab benchmark --model phi3.5 --verbose        # validate a model
python -m onsetlab benchmark --compare phi3.5,qwen2.5:7b    # compare models
python -m onsetlab export --format docker -o ./my-agent      # export as Docker
python -m onsetlab export --format config -o agent.yaml      # export as YAML

Export formats: YAML (portable config), Docker (Dockerfile + compose + Ollama), vLLM (GPU-accelerated), Script (standalone .py file). See Export & Deploy docs for details.

---

Tested Models

| Model | Size | RAM | Notes | |-------|------|-----|-------| | phi3.5 | 3.8B | 4GB+ | Default. Good balance of speed and quality | | qwen2.5:3b | 3B | 4GB+ | Fast, good for simple tasks | | qwen2.5:7b | 7B | 8GB+ | Strong tool calling | | qwen3-a3b | MoE, 3B active | 16GB+ | Best tool calling accuracy | | llama3.2:3b | 3B | 4GB+ | General purpose |

Works with any Ollama model. Run python -m onsetlab benchmark --model your-model to verify.

---

Website · Playground · Documentation · PyPI

Apache 2.0

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: riyanshibohra
• Source: riyanshibohra/OnsetLab
• License: Apache-2.0
• Homepage: https://onsetlab.app/

Install and usage instructions live in the source repository linked above.