Astraeus IDE — Run on the Linux PC

You are at the offline Ubuntu/ROCm machine. The project should live in the current directory (.). Follow these four steps in order.

Step 1 — Activate the venv

The launcher open_IDE expects the venv at ../Agents/venv.

source ../Agents/venv/bin/activate

If the venv doesn't exist yet (first time on this PC), create it:

sudo apt install python3.12 python3.12-venv python3-tk     # one-time, needs net
cd .
python3.12 -m venv venv
source venv/bin/activate

Step 2 — Install the Python deps from the local wheels (offline)

The project ships its own wheels under wheels-linux/. No internet needed:

pip install --no-index \
            --find-links ./wheels-linux \
            -r ./requirements-linux.txt

Quick check that it worked:

python -c "import requests, jinja2; print('ok')"

If pip says "no matching wheel", your Python is not 3.12. Run python3 --version and re-download wheels for that version on the Windows PC.

Step 3 — Launch the IDE

From the project folder:

python main_window.py


Error: (venv) user@hostname:~/IDE$ python main_window.py

Traceback (most recent call last): File "./main_window.py", line 1202, in app = MainWindow(root) ^^^^^^^^^^^^^^^^ File "./main_window.py", line 71, in init self.memory_bridge = MemoryBridge('memory.sqlite3') ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "./memory_bridge.py", line 12, in init self._initialize_global_patterns() File "./memory_bridge.py", line 22, in _initialize_global_patterns cursor = self.conn.cursor() ^^^^^^^^^^^^^^^^ AttributeError: 'NoneType' object has no attribute 'cursor'

Or use the helper:

./open_IDE

A Tk window titled "Astraeus IDE" should open.

Step 4 — Test the AI agent end-to-end

Inside the GUI:

Pick a model — top-right dropdown → Astraeus.
Start Model — click the Start Model button. Logs stream into the terminal pane below; wait for "Server responded to /v1/models".
Agent mode on — make sure the Agent checkbox next to the model dropdown is ticked (it's on by default).
Open a workspace folder — File → Open Folder (or the explorer button) → pick the folder you want the AI to work inside. The agent is locked to that folder.
Chat — in the chat input type something like "list the files here and read main_window.py" and press Send. You should see:
- your message in the chat pane,
- one or more lines like → tool: list_dir(...) and ✓ ok as the agent acts,
- a final Astraeus: ... answer.

If the model never responds

Check the terminal pane for llama-server errors.
Confirm the server is up: curl http://127.0.0.1:8081/v1/models

Run the start command directly in a separate terminal to see raw error output:

export HIP_VISIBLE_DEVICES=0 && \
  ../llama.cpp/build/bin/llama-server \
    -m ../Agents/Astraeus/models/Astraeus_Dolphin_Venice_Full_Hf-24B-F16.gguf \
    -c 32768 --host 127.0.0.1 --port 8081

If llama-server lives somewhere else now, edit the LLAMA_SERVER constant at the top of config.py (one place — it's used by all three model entries).

Things to remember

Never pass -ngl to llama-server on this ROCm box — it will crash. ROCm scales layers automatically.
Only one llama-server at a time (24 GB VRAM cap). The IDE's Start Model button kills any previous one before launching a new one.
The agent has full read / write / run-bash powers, but only inside the workspace folder you opened. Anything outside is refused.
Changing the workspace folder resets the agent's chat history.
The three configured models all share 127.0.0.1:8081, so switching models means stopping the current one and starting another.