Sandbox environments

Give agents a real workspace, not just API calls.

Weavz sandboxes combine hosted browser sessions, code execution, installable packages, Filesystem, and State KV so agents can do the messy work that happens between tools.

workspace/sandboxes/run_7f3
MCP / SDK / API
Agent Browser
Sandbox
Filesystem
State KV
Environment model

Four runtime primitives, one governed workspace

Add only the environments a workflow needs. They inherit workspace integrations, aliases, end-user identity, input partials, Human Gates, and scoped API-key access.

Interactive map

Pick the right sandbox for the step

Agents can mix deterministic tools, browser sessions, and code runs inside one workspace without asking users to reconnect accounts for every environment.

Hosted Chrome

Agent Browser

run.tsworkspace-scoped
await client.actions.execute("agent-browser", "request_human", {
  workspaceId,
  endUserId,
  input: { reason: "MFA required" }
})
viewerUrl ready - agent paused until control returns

Direct automation endpoint for Playwright, Puppeteer, browser-use, and custom drivers

Viewer link for human login without sharing passwords with the agent

End-user scoped browser state so repeat work can continue in the right account

Install what the work needs

From Playwright to pandas, the sandbox can become the toolbench.

Agents can use specialized packages for the parts APIs do not cover: rendering reports, parsing statements, validating files, reconciling exports, or controlling a logged-in browser through the hosted browser session.

Scoped credentials
Human handoff
Approval gates
Retryable state
sandbox install examples
Browser worknpm install playwright browser-use
Data workpip install pandas pyarrow beautifulsoup4
Document workpip install pypdf python-docx openpyxl
Media workuse shell tools for image, PDF, and archive processing
Examples

Workflows that need more than a connector

Operate a legacy supplier portal

Open the vendor dashboard, ask a human to sign in, download invoices, parse PDFs in Sandbox, and save normalized rows to Filesystem.

Investigate customer incidents

Read app dashboards in Browser, pull API data through integrations, run Python analysis, then create a Slack summary gated by approval.

Generate board-ready reporting packs

Install data and document packages, combine CRM and finance exports, render charts, and store the finished PDF for download.

Run QA against logged-in apps

Test critical customer flows in a signed-in browser, capture screenshots, and preserve traces for follow-up runs.

Build stateful research agents

Use Web Reader, Browser, Filesystem, State KV, and Sandbox together so research survives retries, approvals, and handoffs.

Automate internal tools without APIs

Let agents use the same admin UI your team uses while keeping credentials behind human takeover and scoped workspace policy.

Human-safe browser automation

Passwords stay with the person. The signed-in session stays useful.

When a page asks for a password, MFA challenge, or CAPTCHA, Agent Browser returns a viewer link. The human controls the live Chrome session, the agent is paused, and automation resumes only after control is handed back.

Step 1

Agent pauses

Step 2

Human signs in

Step 3

Agent resumes

Add sandbox environments to your agent workspace

Start with Browser, Sandbox, Filesystem, and State KV as first-party workspace tools.