# TryCase

> TryCase gives coding agents a disposable Linux desktop to run, test, and verify apps end-to-end, returning screenshots, recordings, and logs as proof.

TryCase is a cloud-based testing environment platform that lets coding agents spin up disposable Linux machines, run applications, interact with them like a real user, and return verifiable proof — screenshots, video recordings, and logs. It is designed to close the loop between AI-generated code and human review by letting the agent test before the developer does.

## What It Is

TryCase provides on-demand, ephemeral Linux environments that coding agents (Claude, Codex, Cursor, and others) can control via a CLI and a set of installable "skills." Each environment supports terminal sessions, browser automation, filesystem operations, and artifact collection. Environments are destroyed after the task completes, so there is no persistent state to manage. The core workflow is: install skills once, ask the agent for proof on every task, review what comes back.

## How the Workflow Operates

The upload-first model is the recommended starting point. An agent:
1. Installs TryCase skills with a single `npx skills add` command
2. Creates a No Source project and launches a disposable environment
3. Uploads the local working tree with `--respect-gitignore`
4. Runs installs, starts the app, and drives the browser
5. Captures screenshots, recordings, and logs as artifacts
6. Destroys the environment and confirms it is no longer billable

The agent can iterate automatically — the homepage shows a `while (!result.passed)` loop where the agent fixes failures and retests until the flow passes.

## Environment Modes and Sizes

TryCase offers two environment modes:
- **Headless** — default for CLI and LLM verification; supports terminal, browser automation, filesystem, logs, screenshots, and recordings without a visible desktop
- **Desktop** — visible Linux desktop for tasks requiring live viewing, desktop apps, mouse/keyboard APIs, or manual user steps like OAuth consent, CAPTCHA, 2FA, or passkey flows

Four runner sizes are available — nano (1 vCPU, 1 GiB RAM, 10 GiB disk), small (1 vCPU, 2 GiB RAM, 20 GiB disk), standard (2 vCPU, 4 GiB RAM, 40 GiB disk), and large (4 vCPU, 8 GiB RAM, 80 GiB disk) — chosen based on the workload's memory, CPU, and disk needs.

## Agent Integration and Skills System

TryCase ships a universal skills package (`bencsn/trycase-skills`) that teaches coding agents the upload-first route, evidence requirements, and CLI workflow. Once installed, agents can be prompted with natural language like "test the changes end to end and show me a video recording." The skills cover:
- `trycase-cli` — core CLI knowledge and credential management
- `trycase-run-linux-app` — the full upload, run, verify, and destroy loop

The CLI is available via `npx trycase@latest`, `bunx`, or `pnpm dlx`, and supports runtimes including npm, pnpm, yarn, bun, docker-compose, and blank environments.

## Security and Secret Handling

Project secrets are encrypted at rest, scoped to a project, and never returned in plaintext. The platform redacts known secret values from runner output, terminal output, and dashboard errors. Secrets can be imported from local dotenv files (with explicit user approval), set via stdin to avoid shell history exposure, or declared in `trycase.yaml` for reproducible setups. Manual human-only steps — login, CAPTCHA, 2FA, payment confirmation — are handled by switching to desktop mode and giving the user a live take-control link rather than asking for sensitive values in chat.

## Current Status

TryCase is in public beta. The docs reference `large` as "currently the largest public-beta size," and the platform is actively accepting subscribers across multiple tiers. The contact email `ben@trycase.dev` is used for manual credit top-ups and extra credit requests, suggesting a lean early-stage operation.

## Features
- Disposable Linux environments for coding agents
- Headless and desktop environment modes
- Browser automation (click, fill, screenshot, recording)
- Terminal sessions inside environments
- Filesystem upload and artifact collection
- Video recordings and screenshots as proof
- Agent skills package for Claude, Codex, Cursor
- Four runner sizes: nano, small, standard, large
- Project secrets management with encryption at rest
- Secret redaction from all output
- Docker Compose, npm, pnpm, yarn, bun runtime support
- trycase.yaml manifest for reproducible setups
- Automatic environment destruction after task
- Live desktop take-control link for manual steps
- Artifact bundling and download

## Integrations
Claude, Codex, Cursor, npm, pnpm, yarn, bun, Docker Compose, Chromium, Linux

## Platforms
WINDOWS, LINUX, ANDROID, WEB, API, CLI

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://trycase.dev
- Documentation: https://www.trycase.dev/docs
- EveryDev.ai: https://www.everydev.ai/tools/trycase