LoginSign Up
← Back to all sparks
D

Daytona

INFRA · APIS
Velocity5.0

Secure elastic infrastructure for running AI-generated code in isolated sandboxes for AI agents and apps

www.daytona.io ↗

Daytona is shipping a sandbox API every week or two, and GPUs just got cheaper to rent.

dev-sandboxesai-infrastructuresdkgpu-computedeveloper-tools
◆Current state
Daytona releases on a roughly weekly SDK and CLI cadence, each version a small, specific addition to the sandbox control surface. The latest adds warm pool management APIs across all SDKs, spot GPU support, and an OpenTelemetry endpoint override per sandbox. Recent releases have been filling in the operational primitives around sandboxes — snapshots by name, outbound proxy configuration, pre-signed file URLs, typed error codes, enforced TLS.
◆Where it's heading
The direction is toward sandboxes as fleet infrastructure rather than individual dev environments: warm pools, spot capacity, TTLs, auto-pause intervals and metrics are all things you need when something else is provisioning sandboxes in bulk. Error handling has been getting the same treatment — typed codes made consistent across every SDK, which matters for callers that must branch on failure without parsing strings. Fork and snapshot creation graduating to stable in July signals the core lifecycle is considered settled.
◆Prediction
Spot GPU support with warm pools points at scheduling and cost controls next — capacity policies or budget limits are the natural follow-on to renting interruptible hardware. The entries are one-line release summaries linking off-site, so the depth of each change is not readable from the feed alone.

◆Recent moves

  1. 1d ago

    Warm pool management and spot GPUs

    Warm pool management moves into the SDKs, and sandboxes can now run on spot GPUs with a per-sandbox OpenTelemetry endpoint override. Spot capacity is the cost lever for anyone running GPU sandboxes at volume; warm pools are the latency one, and the pairing is consistent with Daytona building for fleets rather than single sessions.

  2. 8d ago

    Snapshot operations by name and outbound proxy

    Snapshots become addressable by name as well as ID, outbound proxy configuration lands at sandbox create, and git transport errors get typed codes across every SDK. Quality-of-life for callers scripting against snapshots, plus the network control an enterprise deployment needs.

  3. 19d ago

    Org members command and client-side HTTP timeout

    Organization member listing arrives in the CLI and the TypeScript SDK gets configurable client-side HTTP timeouts, with daemon error codes made consistent across SDKs. Administrative surface catching up to the runtime surface.

  4. 21d ago

    Stable sandbox fork and snapshot creation

    Sandbox forking and snapshot creation graduate from experimental to stable. A commitment marker rather than a new capability — it tells integrators the core lifecycle API is now safe to build against.

  5. 23d ago

    Pre-signed file URLs and typed SDK errors

    Pre-signed download and upload URLs for sandbox files, typed error codes, and lifecycle-aware sandbox listing. Pre-signed URLs take file transfer off the API path, which is what large artifacts need.

  6. 29d ago

    TLS enforcement and configurable Go SDK timeout

    TLS verification becomes enforced, config file permissions tighten, and the Go SDK gets configurable timeouts. Security defaults being corrected rather than added — the kind of change that only shows up as a break for someone relying on the old leniency.