Mekler AIConnection guide
Put a service to work with your AI.
Choose a service, check what your assistant needs, then get a first result.
New to this? Use the guided first-result task → This page contains manual connection and execution details.
Selecting an operation opens its setup guide.
Model Inference
Generate model completions with automatic task-based model selection, explicit model choice and parallel batches
Example: 0.001939 USDC per completed purchase; example shown, exact quote required
Paid only. No free trial for this operation.
1. Give your assistant the service instructions.
Mekler works through tools your assistant can call. You can use any model with a compatible MCP client or an HTTP tool integration. A chat without those tools cannot run a service on its own.
Help me complete this task with Mekler: Generate model completions with automatic task-based model selection, explicit model choice and parallel batches. Read the agent instructions at https://api.mekler.ai/llms.txt and the exact contract for operation generateModelCompletions at https://api.mekler.ai/catalog.json. Human service details: https://mekler.ai/services/model-inference/. Check current trial eligibility. Explain any required tools or wallet setup. Do not spend funds or accept referral terms without my explicit authorization. Preserve the original input, request ID and proof for retries and recovery.
Your assistant needs web access to read these instructions and execution tools to act on them. This handoff does not install a connection or authorize a purchase.
2. Choose how to connect.
Use an MCP connection
If your assistant supports remote MCP, add Mekler in its tool or connector settings. It can then find services and read their requirements.
Choose your MCP client →Use your own framework
Call the HTTPS API from your agent’s tools. Python, JavaScript and other HTTP-capable environments can use the same service contracts.
Connect through the API →3. Get a result.
Use the existing budgeted inference client: it quotes the exact price, keeps a private purchase journal and recovers uncertain results.
Inference client installation and complete workflow →
Keep the result and recovery information.
Pending or interrupted paid work must be recovered with the original body and proof. Never create a replacement purchase to resolve uncertainty. For trials, retry the identical body and request ID; a completed result can be recovered for seven days.