Feat/ollama delegator agent

[nirabo]

PR entirely generated by Jules/google.

1. USER PROMPT

This repo is a great resource and provides a easy entrypoint to agents.

Let's inspect it and see how we can help make it better.

We always strive to use locally deployed tooling: ollama, n8n, qdrant db, fastapi, bash, linux, minio, nuclio etc.

Let's come up with an improvement to integrate with such self-managed OpenSource services, as plugins to this single-file-agents repo

2. USER PROMPT

modify the PR so that users may run the entire agent locally (via ollama served models) - use gemma3:4b and gemma3:1b from ollama. Also, allow the users to specify their ollama credentials and URLS etc via an .env file. Provide an example template .env initially and let the users setup their backend. Add README entries to detail this process.

3. USER PROMPT

Fantastic work. Check the output generated on my Machine:

lpetrov@tower:~/projects/sandbox/single-file-agents (feat/ollama-delegator-agent)$ uv run sfa_ollama_delegator_agent.py --prompt "warum ist die banane krum" ╭─────────────── Request ────────────────╮ │ User Prompt: warum ist die banane krum │ ╰────────────────────────────────────────╯ Effective Provider: ollama, Primary Model: qwen3:latest, Ollama URL: http://192.168.0.190:11434/, Max Loops: 7 ────────────────────────────────────────────────────────────────── Agent Loop 1/7 (Provider: ollama) ────────────────────────────────────────────────────────────────── [16:38:47] Ollama Chat: Sending request to http://192.168.0.190:11434/api/chat for model 'qwen3:latest'. Payload: { sfa_ollama_delegator_agent.py:183 "model": "qwen3:latest",
"messages": [
{
"role": "user",
"content": "warum ist die banane krum"
}
],
"stream": false,
"options": {
"temperature": 0.7
},
"tools": [
{
"type": "function",
"function": {
"name": "run_ollama_generate",
"description": "Delegates text generation to a specific local Ollama model (e.g., for specialized tasks). The
primary LLM will then analyze or use this output.",
"parameters": {
"type": "object",
"properties": {
"model_name": {
"type": "string",
"description": "The name of the Ollama model to use for this specific task (e.g., 'qwen3:latest'). Must be
available in the local Ollama instance."
},
"prompt": {
"type": "string",
"description": "The specific prompt to send to this delegate Ollama model."
},
"system_prompt": {
"type": "string",
"description": "Optional system prompt for the delegate Ollama model."
},
"context_window": {
"type": "integer",
"description": "Optional context window size for the delegate Ollama model."
}
},
"required": [
"model_name",
"prompt"
]
}
}
}
],
"tool_choice": "auto"
}
[16:39:03] Ollama Chat: Successfully received response from model 'qwen3:latest'. sfa_ollama_delegator_agent.py:203 ╭─────────────────────────────────────────────────────────────────── Agent (ollama - qwen3:latest) ───────────────────────────────────────────────────────────────────╮ │ │ │ │ │ Die Banane ist gekrümmt, weil sie während ihres Wachstums in einer spiraligen Form um den Stängel des Bananenbaums wächst. Dieses Wachstumsmuster ist Teil der │ │ natürlichen Entwicklung des Fruchtknotens, der sich aus den Blüten des Baums bildet. Die Krümmung dient dazu, den Fruchtknoten zu schützen und möglicherweise die │ │ Reife zu fördern. Es ist ein evolutionärer Anpassungsmechanismus, der sicherstellt, dass die Banane sich optimal entwickeln und veredeln kann. │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯ --- Agent delivered final text response. --- Agent run completed successfully.

NOW, let's think of how to finalzie this PR. The template .env.sample must be slightly changed, so that the primary LLM provider can be of a broader range (see initial .env.sample file.

[nirabo]

Title: Feat: Enable Fully Local Ollama Execution & .env Config for Ollama Delegator Agent

Body:

Summary
This PR significantly enhances the sfa_ollama_delegator_agent.py by enabling fully local LLM execution using Ollama for the primary agent logic and introducing robust configuration via .env files. These changes were generated by me, Jules, an AI agent.

Key Features and Changes:
Fully Local Primary LLM with Ollama:

The agent can now be configured to use an Ollama-hosted model (e.g., gemma2:9b, llama3) as its primary orchestrating LLM, removing the dependency on cloud-based providers like Anthropic if desired.
A new get_ollama_chat_completion function was added to handle interactions with Ollama's /api/chat endpoint, including support for tool calls.
.env File Configuration:

Integrated python-dotenv for easy configuration management.
You can now define settings such as PRIMARY_LLM_PROVIDER, Ollama URLs, default model names for primary and delegate roles, and Anthropic API keys/models in a .env file.
A comprehensive .env.sample file is provided as a template, with clear explanations for each variable.
CLI arguments remain functional and will override .env settings.
Enhanced Agent Logic:

The main agent loop in sfa_ollama_delegator_agent.py now dynamically switches between Anthropic and Ollama based on the chosen configuration.
Tool use (specifically the run_ollama_generate tool for delegating tasks to another Ollama model) is handled consistently across both primary LLM providers.
Updated System Prompt:

The AGENT_SYSTEM_PROMPT has been generalized to be suitable whether the primary LLM is Claude or an Ollama-hosted model.
Comprehensive Documentation:

The README.md has been extensively updated to detail these new capabilities, including:
How to set up and use the .env file.
Instructions for running the agent with Ollama as the primary LLM.
Updated command-line argument descriptions.
Example usage scenarios for both Ollama and Anthropic primary LLM configurations.
Benefits:
Flexibility: You can choose between a fully local setup (Ollama for both primary and delegate roles) or a hybrid approach (e.g., Anthropic primary, Ollama delegate).
Privacy & Cost: Enables use in air-gapped environments or to leverage local models for cost savings.
Ease of Configuration: Simplifies setup through the .env file.
This set of changes makes the Ollama Delegator Agent more versatile and user-friendly, particularly for those prioritizing local LLM solutions.

This Pull Request was generated by me, Jules (AI Agent).

Let me know if you'd like any adjustments to this!