Instructions to use renatoschust/gama-agent with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use renatoschust/gama-agent with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf renatoschust/gama-agent:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf renatoschust/gama-agent:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf renatoschust/gama-agent:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf renatoschust/gama-agent:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf renatoschust/gama-agent:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf renatoschust/gama-agent:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf renatoschust/gama-agent:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf renatoschust/gama-agent:Q4_K_M

Use Docker

docker model run hf.co/renatoschust/gama-agent:Q4_K_M

LM Studio
Jan

vLLM

How to use renatoschust/gama-agent with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "renatoschust/gama-agent"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "renatoschust/gama-agent",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/renatoschust/gama-agent:Q4_K_M

Ollama
How to use renatoschust/gama-agent with Ollama:
```
ollama run hf.co/renatoschust/gama-agent:Q4_K_M
```
Unsloth Desktop

How to use renatoschust/gama-agent with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf renatoschust/gama-agent:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "renatoschust/gama-agent:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use renatoschust/gama-agent with Docker Model Runner:
```
docker model run hf.co/renatoschust/gama-agent:Q4_K_M
```

Lemonade

How to use renatoschust/gama-agent with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull renatoschust/gama-agent:Q4_K_M

Run and chat with the model

lemonade run user.gama-agent-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use renatoschust/gama-agent with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf renatoschust/gama-agent:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default renatoschust/gama-agent:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use renatoschust/gama-agent with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf renatoschust/gama-agent:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "renatoschust/gama-agent:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

GAMA Agent 35B

GAMA Agent 35B is an open-source agent / tool-calling language model for Bosnian (and regional HR/SR), built as a fine-tune of Qwen3.5-35B-A3B (Mixture-of-Experts: ~35B total, ~3B active). It is tuned to act, not just chat: given a task and a list of tools, it emits structured tool calls so it can drive software, query databases, run commands and complete multi-step jobs.

Built by gamAI — an open AI toolkit. Published as a ready-to-run GGUF (Q4_K_M, ~21 GB) quantisation, so it runs on a single mid-range GPU with CPU offload, or on CPU with enough RAM.

What it is


Base model	`Qwen/Qwen3.5-35B-A3B` (MoE, ~35B total / ~3B active)
Fine-tune	SFT + DPO agent tuning — LoRA r=16, α=16
Format	GGUF `Q4_K_M` (~21 GB)
Languages	Bosnian (primary), Croatian, Serbian, English
Context	up to 32k tokens
Adapter	`renatoschust/gama-agent-lora`

What it can do

Bosnian conversation — fluent, natural (ijekavica/latinica).
Tool / function calling — returns structured tool calls (JSON) for tools such as: shell/terminal, file read/write, service status, SQL/database queries, HTTP/web fetch, calculations.
Multi-step agent tasks — diagnose → act → verify loops from a single instruction.
Safety — refuses destructive actions (e.g. rm -rf, dropping a database, deleting logs) instead of executing them.
Non-thinking — direct answers, low latency (no long reasoning preamble).

Quickstart

Ollama

ollama run hf.co/renatoschust/gama-agent

llama.cpp (llama-server, OpenAI-compatible API)

llama-server -hf renatoschust/gama-agent -c 32768 --port 8080

Then call it like any OpenAI endpoint:

curl http://127.0.0.1:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
  "messages":[{"role":"user","content":"Koliko je 15% od 240?"}],
  "temperature":0
}'

LM Studio / GPT4All

Download the GGUF file and load it directly.

Tool-calling format

Pass your tools in the standard OpenAI tools / functions schema and use a low temperature (0–0.2). Example system prompt:

Ti si asistent. Za svaki zahtjev koji traži stanje sistema ili akciju koristi odgovarajući alat (function call).
Ako zahtjev traži opasnu ili razornu akciju (brisanje, formatiranje, gašenje), ODBIJ i ne pozivaj alat.
Ako je pitanje opće prirode, odgovori normalno bez poziva alata.

Prompt template (chat template)

The model uses the Qwen3.5 chat template (see chat_template.jinja in the adapter repo). If you run it via Ollama/llama.cpp with the GGUF, the embedded template is used automatically.

Training data

Trained on synthetic, placeholder-sanitised Bosnian agent traces (function-calling, terminal/ops trajectories, confirmation/rebalance and DPO preference pairs). No real customer, company or personal data was used.

Limitations

Bosnian-first — quality drops on other languages.
~21 GB — needs a machine with sufficient RAM/VRAM (a 24 GB GPU or ≥24 GB RAM with offload).
Agent reliability depends on the tool schema and system prompt you provide.
Not a knowledge base — for facts, ground it with retrieval (RAG).
Quality may vary across rare/ambiguous tasks; always keep a human gate on destructive operations.

License

Apache-2.0 (inherited from the Qwen3.5-35B-A3B base model). You are responsible for how you use it.

Built with ❤️ by gamAI — open AI toolkit · GitHub

Downloads last month: 19

GGUF

Model size

35B params

Architecture

qwen35moe

Hardware compatibility

4-bit

Model tree for renatoschust/gama-agent

Base model

Qwen/Qwen3.5-35B-A3B-Base

Finetuned

Qwen/Qwen3.5-35B-A3B

Quantized

(302)

this model