Skip to main content
The join command connects your local LLM endpoint to a Gambiarra room, making your models available to other participants through the hub.

​
Usage

gambiarra join --code <room-code> --model <model-name> [options]

​
Options

​
--code
string
required
Room code to join. Also accepts -c shorthand. This is the code you received when the room was created.
gambiarra join --code ABC123 --model llama3
​
--model
string
required
Model to expose from your endpoint. Also accepts -m shorthand. Must be available on your LLM server.
gambiarra join --code ABC123 --model llama3
​
--endpoint
string
default:"http://localhost:11434"
OpenAI-compatible API endpoint. Also accepts -e shorthand. Works with Ollama, LM Studio, and other OpenAI-compatible servers.
gambiarra join --code ABC123 --model llama3 --endpoint http://localhost:11434
​
--nickname
string
Display name for your endpoint. Also accepts -n shorthand. If not provided, defaults to <model>@<id-prefix>.
gambiarra join --code ABC123 --model llama3 --nickname "My GPU Server"
​
--password
string
Room password (if the room is password-protected). Also accepts -p shorthand.
gambiarra join --code ABC123 --model llama3 --password secret123
​
--hub
string
default:"http://localhost:3000"
Hub URL to connect to. Also accepts -H shorthand.
gambiarra join --code ABC123 --model llama3 --hub http://192.168.1.10:3000
​
--no-specs
boolean
default:"false"
Don’t share machine specs (CPU, RAM, GPU) with the room. By default, the command detects and shares your system specifications.
gambiarra join --code ABC123 --model llama3 --no-specs

​
Examples

​
Join with Ollama

Connect your local Ollama server to a room: 
gambiarra join --code ABC123 --model llama3 --endpoint http://localhost:11434
Output: 
Detected specs: Intel i7-9700K, 32GB RAM, NVIDIA RTX 3080

Joined room ABC123!
  Participant ID: xK8pQ2mN5v
  Nickname: llama3@xK8pQ2
  Model: llama3
  Endpoint: http://localhost:11434

Your endpoint is now available through the hub.
Press Ctrl+C to leave the room.
The command stays running and sends periodic health checks to the hub. Keep the terminal open to remain in the room.

​
Join with LM Studio

LM Studio typically runs on port 1234: 
gambiarra join --code ABC123 --model gpt-4 --endpoint http://localhost:1234

​
Join with custom nickname

Set a friendly display name: 
gambiarra join --code ABC123 --model llama3 --nickname "Office GPU"
Output: 
Detected specs: Intel i7-9700K, 32GB RAM, NVIDIA RTX 3080

Joined room ABC123!
  Participant ID: aB3cD4eF5g
  Nickname: Office GPU
  Model: llama3
  Endpoint: http://localhost:11434

Your endpoint is now available through the hub.
Press Ctrl+C to leave the room.

​
Join password-protected room

Provide the room password: 
gambiarra join --code ABC123 --model llama3 --password secret123

​
Join remote hub

Connect to a hub on a different machine: 
gambiarra join --code ABC123 --model llama3 --hub http://192.168.1.10:3000

​
Hide system specs

Prevent sharing your hardware information: 
gambiarra join --code ABC123 --model llama3 --no-specs

​
Complete example with all options

gambiarra join \
  --code ABC123 \
  --model llama3 \
  --endpoint http://localhost:11434 \
  --nickname "Main Server" \
  --password secret123 \
  --hub http://192.168.1.10:3000

​
How it works

  1. Model validation: The command connects to your endpoint and verifies the model exists
  2. Registration: Sends a join request to the hub with your model, endpoint, and specs
  3. Health checks: Sends periodic health checks every 10 seconds (configurable in core)
  4. Active session: Remains connected until you press Ctrl+C or connection is lost
  5. Graceful exit: Notifies the hub when leaving
Health checks run every 10 seconds. If 3 consecutive checks fail (30 seconds), you’re automatically removed from the room.

​
Supported endpoints

Gambiarra works with any OpenAI-compatible API server:
  • Ollama: Default endpoint http://localhost:11434
  • LM Studio: Default endpoint http://localhost:1234
  • LocalAI: Configurable endpoint
  • Custom servers: Any server implementing OpenAI’s API format

​
Endpoint detection

The command automatically detects your endpoint type:
  1. Tries Ollama API format (/api/tags)
  2. Falls back to OpenAI format (/v1/models)
  3. Lists available models from the detected format

​
Error handling

​
Model not found

If the specified model isn’t available: 
gambiarra join --code ABC123 --model nonexistent
Output: 
Model 'nonexistent' not found.
Available models: llama3, llama2, codellama

​
No models at endpoint

If the endpoint has no models: 
gambiarra join --code ABC123 --model llama3 --endpoint http://localhost:9999
Output: 
No models found at http://localhost:9999
Make sure your LLM server is running and has models available.

​
Wrong password

If the password is incorrect: 
Error: Invalid password

​
Room not found

If the room code doesn’t exist: 
Error: Room not found

​
Connection lost

If connection to the hub is lost during the session: 
Lost connection to hub, leaving room...
The command exits automatically.

​
Health check behavior

While connected:
  • Health checks are sent every 10 seconds
  • Participant timeout is 30 seconds (3 failed checks)
  • If health checks fail, you’re removed from the room
  • Graceful shutdown notifies the hub immediately

​
System specs detection

By default, the command detects and shares:
  • CPU: Model and core count
  • RAM: Total system memory
  • GPU: Graphics card model (if available)
Disable with --no-specs if you prefer not to share.
  • create - Create a new room
  • list - List available rooms
  • monitor - Monitor room activity in real-time
  • serve - Start the hub server

​
Next steps

After joining a room:
  1. Keep the terminal open to stay connected
  2. Monitor the room from another terminal
  3. Other participants can now route requests to your model
  4. Press Ctrl+C when you want to leave

Build docs developers (and LLMs) love

Get started for freeTalk to us