Architecture

PandaStack has three main layers:

1. Control plane — API, auth, organizations, tokens, templates, billing, and orchestration.
2. Data plane — agents that run sandbox operations on hosts.
3. Workload microVMs — Firecracker sandboxes created from templates.

Public surfaces

• Marketing: pandastack.ai
• Dashboard: app.pandastack.ai
• API: api.pandastack.ai
• Docs: docs.pandastack.ai

API entry points

GET /healthz
GET /version
GET /metrics
GET /v1/me
POST /v1/sandboxes
GET /v1/templates
POST /v1/billing/sync

Sandbox operations are scoped under /v1/sandboxes/{id}. Filesystem access uses /fs, lifecycle uses routes such as /pause, /resume, /snapshots, /fork, /hibernate, and /wake, and streaming uses /exec/stream, /events, and WebSocket upgrades.

Agent-proxied operations

POST /v1/sandboxes/{id}/exec
POST /v1/sandboxes/{id}/exec/stream
GET /v1/sandboxes/{id}/exec/pty
GET /v1/sandboxes/{id}/exec/ws
GET /v1/sandboxes/{id}/logs
GET /v1/sandboxes/{id}/events
GET /v1/sandboxes/{id}/lsp/{lang}

Client configuration

export PANDASTACK_API_KEY=pds_abc123def456...
export PANDASTACK_API=https://api.pandastack.ai

The Python SDK imports from pandastack; the TypeScript SDK imports from @pandastack/sdk; the CLI binary is pandastack.

Lifecycle model

Use Sandbox.create(...) or pandastack sandbox create to start a sandbox, exec or run_code to run work, sandbox.filesystem or sandbox cp for files, and kill or pandastack sandbox delete to clean up. Advanced lifecycle operations include pause, resume, snapshot, fork, hibernate, wake, TTL changes, and persistence changes.