Documentation

Klerix in 10 minutes.

A typed workflow graph, a sandboxed runtime, a verifier ensemble. Everything you need to ship your first agent into production by Friday.

1. Install & auth

Install the Klerix CLI. Authenticate with your workspace token from the dashboard.

# macOS / linux
curl -fsSL https://klerix.com/install.sh | sh

# authenticate
klerix auth login --token $KLERIX_TOKEN

2. Hello, agent

Create a workflow that reads a payer remit and posts the line items to your CRM.

# workflows/post-remit.klx
workflow "post-remit" {
  trigger: schedule("*/10 * * * *")
  use: [availity, salesforce]

  remit = availity.fetch_remit(since: "10m")
  for r in remit {
    salesforce.upsert(object: "Remit__c", data: r)
  }
}

3. Capture a screencast

For portals without a structured API, record a 10-minute screencast. Klerix compiles a workflow graph from the recording.

klerix capture --portal "carrier-X" --out workflows/carrier-x.klx

Workflows

Workflows are typed graphs of step nodes. Each step has retries, an idempotency key, and a verifier policy.

Tools

Tools are typed functions that an agent can invoke. They are versioned and tested. The standard library ships with 200+ portal connectors.

Verifier ensemble

Every consequential action passes through an independent verifier. Below threshold actions escalate to a human reviewer.

Audit trail

Every step writes an immutable log entry: timestamp, input, output, verifier score, model version, screenshot.

Deploy & rollback

klerix deploy ships a new version. Klerix runs it in shadow mode against the prior version's traffic before cut-over.

Evals

Each workflow has a golden set. PRs cannot land if regressions exceed your configured threshold.

Observability

OpenTelemetry traces. Per-step latency, cost, and accuracy. Stream to Datadog, Honeycomb, or your warehouse.