Assay

Assay is a self-hosted telemetry server for Laravel applications that call AI models through laravel/ai. Each of your applications reports what its agents and AI operations did: which provider and model they called, how many tokens and other usage units they consumed, how long each step took, which tool calls ran and which runs failed. Assay stores that record in your own Laravel Cloud account, shows it on a usage dashboard, and lets you turn selected production runs into versioned JSONL datasets.

Assay is Experimental in the Scalpels catalog. Installing it through Scalpels is open to supporters and partner teams.

What Assay does and does not do

• It records usage by application, environment, operation, agent, provider, model and an opaque subject you choose (for example user:123). Usage is always recorded and never sampled.
• Prompt and response content is a separate, opt-in capture mode. By default nothing your users typed and nothing a model returned leaves the reporting application.
• It links agents, steps, attempts, tool calls and sub-agents into one run tree, so you can see what happened inside a single request.
• It lets people with content access label, rate and annotate runs, collect them into datasets, and download a dataset as JSONL on demand.
• It can erase a subject from stored content while keeping usage totals under a pseudonymous tombstone.
• It exposes all of this to AI agents through an MCP server, with usage and content as separate abilities.

Assay reports usage, not money. It ships no model prices, does not reconcile invoices, and does not replay, score or evaluate runs. It performs no redaction or PII detection: if you capture content, your application decides what to mask or drop before it is sent. The reporting side supports Laravel 13 with laravel/ai 1.x only.

How it fits with Scalpels

Scalpels forks artisan-build/assay into your GitHub account, creates the Postgres database, managed queue, object storage bucket and scheduler Assay declares on your Laravel Cloud account, generates Assay's dedicated erasure key, runs the migrations and deploys. See Install a product for how installs work in general.

Assay is a Scalpels provider, so Scalpels can attach a Laravel Cloud environment or Laravel Forge site to it, mint an ingest credential, configure the application and open the client-install pull request. Your agent can reach Assay's MCP tools through the Scalpels MCP server.

In these docs

• Getting started: install Assay through Scalpels, what the install creates, and how to sign in.
• Connecting your apps: install the client package in a reporting application, issue its credential and confirm data is arriving.
• Capturing content: usage and full capture modes, sampling, failure capture, subjects and payload filters.
• The usage dashboard: the dashboard's tables, CSV and JSON downloads, run trees and the risk guide.
• Curation and datasets: flag runs, build datasets and export them as JSONL.
• Access, retention and erasure: who can see content, how long data is kept, erasing a subject and restoring a backup.
• Configuration: the server settings you can change and the values Scalpels manages.
• Use Assay from your agent: Assay's MCP endpoints, tools and credentials.
• Troubleshooting: missing data, rejected envelopes, export and access problems.

Want a branded address for this deployment? Use your own domain.