ashishpatel26/Agent-harness-using-Agno

A step-by-step tour of building AI agents with Agno.
Each script adds one harness capability on top of the last. Every example runs on a local model, with no API keys needed.

A model on its own only predicts text. The harness is everything around it that turns it into an agent: tools, schemas, storage, memory, state, knowledge, guardrails, human approval, and coordination with other agents. This repo builds that harness one layer at a time.

flowchart LR
    U([User]) --> G[Guardrails]
    G --> A{{Agent<br/>qwen via Ollama}}
    A <--> T[Tools<br/>YFinance]
    A <--> K[(Knowledge<br/>ChromaDB)]
    A <--> M[(Storage · Memory · State<br/>SQLite)]
    A -. approval .-> H([Human])
    A --> O([Typed output])

1. Requirements: uv, Ollama, and Python 3.14 (uv installs it for you).

2. Pull the models

ollama pull qwen3.5:4b          # chat model
ollama pull nomic-embed-text    # embeddings for examples 7 and 8

The scripts use the tag qwen3.5-4b-16k, which is qwen with a 16k context window. Create it once:

printf 'FROM qwen3.5:4b\nPARAMETER num_ctx 16384\n' > Modelfile
ollama create qwen3.5-4b-16k -f Modelfile

Using a different model? Change Ollama(id="...") in the scripts.

3. Install and run

git clone https://github.com/ashishpatel26/Agent-harness-using-Agno.git
cd Agent-harness-using-Agno
uv sync
uv run 1-Agent-with-tools.py

Run the examples in order: each one builds on the ideas of the previous script.

.
├── 1-…13-*.py          # the examples, in learning order
├── data/               # source documents for the knowledge base
├── db/                 # SQLite + ChromaDB files (created at runtime, git-ignored)
├── assets/banner.svg
└── pyproject.toml

Agno (agents, teams, workflows, AgentOS) · Ollama (local LLM and embeddings) · ChromaDB (vector store) · SQLite (sessions, memory, state) · yfinance (market data) · Pydantic · FastAPI

The finance examples are for learning only. They are not investment advice, and market data from Yahoo Finance can be delayed or incomplete.

If this helped you learn agent engineering, consider giving it a ⭐