Set Up Pipelex Inference Configuration

Guided setup for configuring the Pipelex runtime with inference backends and API keys. This is only needed to run methods with live inference — building, validating, editing, and dry-running work without any configuration.

Process

Step 0 — CLI Check (mandatory, do this FIRST)

Run mthds-agent --version. The minimum required version is 0.1.2 (declared in this skill's front matter as min_mthds_version).

• If the command is not found: STOP. Do not proceed. Tell the user:

The mthds-agent CLI is required but not installed. Install it with:

npm install -g mthds

Then re-run this skill.

• If the version is below 0.1.2: STOP. Do not proceed. Tell the user:

This skill requires mthds-agent version 0.1.2 or higher (found X.Y.Z). Upgrade with:

npm install -g mthds@latest

Then re-run this skill.

• If the version is 0.1.2 or higher: proceed to the next step.

Step 1 — Check Pipelex Runtime

Run pipelex-agent --version to check if the Pipelex runtime is installed.

• If not installed: Tell the user:

The Pipelex runtime is not installed. Install it with:

curl -fsSL https://pipelex.com/install.sh | sh

Or alternatively: pip install pipelex

Wait for confirmation, then proceed.

Step 2 — Assess Current State

Run mthds-agent pipelex doctor to check the current configuration health.

• If healthy (backends configured, credentials valid): Tell the user their setup is already complete. Suggest trying /run on a method.
• If issues found: Show the doctor output and proceed to Step 3 to fix them.

Step 3 — Choose Backends

Ask the user which backends they want to configure. Two paths:

Option A — Pipelex Gateway (simplest)

The Pipelex Gateway provides access to multiple AI models through a single API key.

1. Ask the user: "Do you accept the Pipelex Gateway Terms of Service and Privacy Policy? See: https://www.pipelex.com/privacy-policy"
2. If they accept: run mthds-agent pipelex login to log in to Pipelex and configure the PIPELEX_GATEWAY_API_KEY. This command handles authentication and key setup automatically.
3. If they decline: proceed with Option B instead

Option B — Bring Your Own Key (BYOK)

The user provides their own API keys for individual providers.

1. Ask which providers they want to enable. Common options: openai, anthropic. Run mthds-agent pipelex init --help to discover all available backends.
2. If 2+ backends are selected, ask which should be the primary_backend.

Step 4 — Telemetry Preference

Ask the user: "Do you want anonymous telemetry enabled?"

• "off" — no telemetry
• "anonymous" — anonymous usage data

Step 5 — Apply Configuration

You MUST have all answers from Steps 3-4 before running this command.

Build the JSON config and run:

# Example: Pipelex Gateway (user accepted terms):
mthds-agent pipelex init -g --config '{"backends": ["pipelex_gateway"], "accept_gateway_terms": true, "telemetry_mode": "anonymous"}'

# Example: BYOK with OpenAI only:
mthds-agent pipelex init -g --config '{"backends": ["openai"], "telemetry_mode": "off"}'

# Example: BYOK with multiple backends:
mthds-agent pipelex init -g --config '{"backends": ["openai", "anthropic"], "primary_backend": "anthropic", "telemetry_mode": "off"}'

• -g targets the global ~/.pipelex/ directory. Without it, targets project-level .pipelex/ (requires a project root).
• Config JSON schema: {"backends": list[str], "primary_backend": str, "accept_gateway_terms": bool, "telemetry_mode": str}. All fields optional.
• When pipelex_gateway is in backends, accept_gateway_terms must be set.
• When 2+ backends are selected without pipelex_gateway, primary_backend is required.

Step 6 — Set API Keys (BYOK only)

If the user chose BYOK (Option B), guide them to set environment variables for their chosen backends:

• OpenAI: export OPENAI_API_KEY="sk-..."
• Anthropic: export ANTHROPIC_API_KEY="sk-ant-..."

Recommend adding these to their shell profile (~/.zshrc, ~/.bashrc, etc.) for persistence.

After the user confirms they've set the keys, run mthds-agent pipelex doctor again to verify everything is healthy.

Step 7 — Confirm Success

Once the doctor reports a healthy configuration:

1. Confirm the setup is complete
2. Suggest trying /run on a method to test the configuration

Reference

• Error Handling — read when CLI returns an error to determine recovery
• MTHDS Agent Guide — read for CLI command syntax or output format details