Worker Network

Community Worker Network Suspended

The BazaarLink Worker Network lets anyone contribute compute — a local GPU, a home server, or a cloud VM — and earn credit rewards for serving inference requests. Community models appear alongside cloud models in the API, priced below standard rates.

Calling Community Models

Community models use the prefix community/ in the model ID. The API is fully OpenAI-compatible — just change the model name.

List available community models

Community models appear in the model list with IDs starting with community/.

curl https://bazaarlink.ai/api/v1/models \
  -H "Authorization: Bearer sk-bl-YOUR_API_KEY" \
  | jq '.data[] | select(.id | startswith("community/"))'

Make a chat completion request

curl https://bazaarlink.ai/api/v1/chat/completions \
  -H "Authorization: Bearer sk-bl-YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "community/qwen2.5:2b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Streaming

Community models support streaming responses exactly like cloud models:

response = client.chat.completions.create(
    model="community/qwen2.5:2b",
    messages=[{"role": "user", "content": "Count to 10 slowly"}],
    stream=True,
)
for chunk in response:
    delta = chunk.choices[0].delta.content or ""
    print(delta, end="", flush=True)

Run a Worker Node

Contribute compute to the network in three steps: install Ollama, install the worker CLI, then connect to the gateway.

Step 1 — Prerequisites

# Pull the model you plan to serve
ollama pull qwen2.5:2b

# Verify Ollama is running
ollama list

Step 2 — Install the worker CLI

npm install -g bazaarlink-worker

# Or run directly without installing
npx bazaarlink-worker --help

Step 3 — Register

Register your email with the platform. After email verification, a worker key (wk_...) is issued automatically and emailed to you.

bazaarlink-worker register \
  --email [email protected] \
  --models qwen2.5:2b \
  --max-concurrent 4

Step 4 — Login with your worker key

Use the worker key (wk_...) from your verification email to authenticate and set your prices:

bazaarlink-worker login \
  --key wk_YOUR_WORKER_KEY \
  --gateway wss://gateway.bazaarlink.ai \
  --models qwen2.5:2b \
  --ollama-url http://localhost:11434 \
  --input-price 0.10 \
  --output-price 0.20

Step 5 — Start serving

bazaarlink-worker start

The worker connects to the gateway via WebSocket and begins accepting inference jobs. Keep the process running (use a process manager like PM2 or systemd for production).

# Check status
bazaarlink-worker status

# Run as background service with PM2
npm install -g pm2
pm2 start "bazaarlink-worker start" --name bazaarlink-worker
pm2 save

Pricing & Rewards

Community model pricing has two layers: the platform price charged to users, and the reward paid to the worker. Both are set in USD per million tokens.

The platform keeps the difference (margin). Your price must not exceed the platform ceiling. If your price exceeds this ceiling, the gateway will reject the connection with an error.

Example

Platform price:   $0.005 / 1M prompt tokens
Max worker price: $0.004 / 1M prompt tokens  (set by platform ceiling)

You set --input-price 0.004  ✓ accepted
You set --input-price 0.005  ✗ rejected (exceeds ceiling)

How rewards are paid

Rewards are credited to your BazaarLink account balance for each completed request. The reward amount equals your price × tokens served. You can spend accumulated credits on any BazaarLink API call or withdraw (future feature).

Data & Privacy

When your node is connected, it processes API requests from BazaarLink users. Understanding what data flows through your node is important before you join.

What passes through your node

• User prompt messages and conversation history for each request assigned to your node.
• Model name and generation parameters (temperature, max tokens, etc.).
• Token counts reported back to the platform after each job.

What does NOT pass through your node

• User identity, API keys, or account information — these are never sent to workers.
• Payment or billing data.
• Other users' requests — each job is dispatched individually.

Platform review & content filtering

The platform applies inbound and outbound content filters on every community request. Prompts containing detected secrets (API keys, tokens) or prompt-injection patterns may be blocked before reaching your node. Responses are similarly scanned before being forwarded to the user.

Provider verification

Before a node joins the Worker Network, the platform performs multi-layer verification: hardware model and resource checks, followed by AI-powered quality probing (Speed Probe + Quality Probe) to ensure the node serves accurate and responsive model outputs. Nodes that fail probing do not receive live traffic.

Your obligations as a node operator

• Do not log or store user prompt content beyond what Ollama requires for inference.
• Do not use user prompts for training or analysis.
• Run your node only on hardware you control and trust.
• Report any suspected data issues to the platform administrator.