Hawk Agent — Setup Instructions

1. Install & Launch

1

Download HawkAgent.exe and place it in its own folder (e.g. C:\HawkAgent\).

2

Double-click HawkAgent.exe. The app opens maximized and lands on a new session, ready to go.

2. Configure the LLM (in-app)

Everything is done from the settings gear (⚙) in the top bar — no text editor needed. Open it and fill in:

Provider

One of: openai-compatible, openai, or openrouter. Use openai-compatible for DeepSeek, local models, and most third-party gateways.

Model

The model name your provider expects, e.g. gpt-4o-mini, deepseek-chat, meta-llama/llama-3.1-8b-instruct:free.

Base URL

The API endpoint. Examples:

https://api.openai.com/v1          # OpenAI
https://api.deepseek.com/v1        # DeepSeek
https://openrouter.ai/api/v1       # OpenRouter
http://localhost:11434/v1         # Ollama (local)

API Key

Your secret key. Leave empty only if the endpoint needs no auth (e.g. a local Ollama with auth disabled). Settings saves it for you.

Settings writes your choices back automatically — they persist between launches.

3. Free API Keys & Free Models

You don't need to pay. These providers offer free tiers or free models — sign up, then paste the key + base URL + model into Settings:

OpenRouter — free models with a :free suffix (e.g. meta-llama/llama-3.1-8b-instruct:free, google/gemma-2-9b-it:free). base_url: https://openrouter.ai/api/v1
Groq — very fast free tier (Llama, Mixtral). base_url: https://api.groq.com/openai/v1
Google AI Studio — free Gemini API key. base_url: https://generativelanguage.googleapis.com/v1beta/openai/
Ollama — run models locally, 100% free, no key needed. base_url: http://localhost:11434/v1, api_key: leave empty
Hugging Face — free inference API for many open models. base_url: https://api-inference.huggingface.co/v1/
Mistral — free tier for small models. base_url: https://api.mistral.ai/v1

Tip: start with OpenRouter's :free models — no credit card, just an account.

4. Optional Tuning (in-app)

temperature — 0.0 to 1.0 (default 0.2). Higher = more creative.
max_tokens — max reply length (default 4096).
request_timeout — seconds per HTTP request (default 120).
max_retries / retry_backoff — resilience for flaky networks.
agent.max_steps — max tool-use steps per task (default 20).
agent.tool_timeout — seconds per tool call (default 30).
tools.filesystem.base_dir — sandbox root for file tools (default .).
tools.shell.allowlist — commands the agent may run.
tools.require_approval — set true to confirm writes/commands before running.

5. Troubleshooting

If you see "No API key set", open Settings and paste your key, then save. Make sure the model and base URL match the provider you chose.