Appearance

Changelog ​

All notable changes to SandBase are documented here. Dates are in YYYY-MM-DD format.

v0.5.4 — 2025-07-01 ​

Added ​

  • Sandbox auto-keepalive — Sandboxes now support automatic keepalive via the POST /sandboxes/:id/refreshes endpoint
  • Sandbox timeout update — Change timeout on running sandboxes with POST /sandboxes/:id/timeout
  • Dashboard sandbox management — View, pause, resume, and stop sandboxes from the dashboard

Fixed ​

  • Fixed sandbox status not updating in real-time on the dashboard
  • Fixed edge case where paused sandboxes could not be resumed after 30+ minutes

v0.5.3 — 2025-06-15 ​

Added ​

  • Seed 1.6 model — ByteDance's multimodal model now available
  • Qwen3-32B — Alibaba's efficient open model added to the roster
  • Llama 4 Maverick — Meta's latest open-source model with vision support

Improved ​

  • Reduced sandbox creation time from ~5s to ~2s
  • Better error messages for capability mismatches

v0.5.2 — 2025-06-01 ​

Added ​

  • Sandbox snapshots — Save and restore sandbox state with POST /sandboxes/:id/snapshots
  • Network configuration — Control outbound network access per sandbox
  • File upload/download — Direct file operations via POST /sandboxes/:id/files

Improved ​

  • Sandbox exec now supports streaming stdout/stderr in real-time
  • Increased max file upload size to 100 MB

v0.5.1 — 2025-05-15 ​

Added ​

  • Desktop template — Full GUI environment with VNC access for browser automation
  • Sandbox pause/resume — Save sandbox state and resume later without losing progress

Fixed ​

  • Fixed timeout not resetting after exec commands
  • Fixed sandbox metadata not persisting across pause/resume cycles

v0.5.0 — 2025-05-01 ​

Added ​

  • Sandbox API — E2B-compatible sandbox environments for code execution and AI agents
  • 7 pre-built templates — code_interpreter, claude, codex, hermes, opencode, desktop, base
  • Sandbox lifecycle management — Create, pause, resume, and destroy sandboxes
  • Process execution — Run commands in sandboxes with POST /sandboxes/:id/processes
  • File system access — Read, write, list, and delete files in sandboxes

Breaking Changes ​

  • None — sandbox API is a new addition

v0.4.5 — 2025-04-20 ​

Added ​

  • Gemini 2.5 Flash — Google's fastest model with 1M context window
  • Gemini 2.5 Pro — Google's most capable model with thinking support
  • DeepSeek Reasoner — Chain-of-thought reasoning model

Improved ​

  • Streaming latency reduced by 15% for Anthropic models
  • Better handling of Anthropic thinking/reasoning blocks in OpenAI format

v0.4.4 — 2025-04-10 ​

Added ​

  • Claude Sonnet 4 — Anthropic's latest model with extended thinking
  • Prompt caching — Support for Anthropic explicit caching and OpenAI automatic caching
  • Cache usage reporting — cached_tokens and cache_creation_tokens in usage response

Improved ​

  • Reduced TTFT (Time-To-First-Token) by 30ms on average
  • Better error messages for authentication failures

v0.4.3 — 2025-03-25 ​

Added ​

  • o3 and o3-mini — OpenAI's reasoning models with reasoning_effort parameter
  • Reasoning token tracking — reasoning_tokens reported in usage
  • Capability-aware routing — Requests are automatically routed to providers that support required capabilities

Breaking Changes ​

  • Requests requiring unsupported capabilities now return 400 capability_unsupported instead of silently dropping parameters

v0.4.2 — 2025-03-10 ​

Added ​

  • Anthropic Messages API — Full /v1/messages endpoint compatibility
  • Anthropic streaming — All 6 event types (message_start, content_block_start/delta/stop, message_delta, message_stop)
  • Cross-protocol routing — Use Anthropic SDK to access OpenAI models and vice versa

Improved ​

  • Tool call streaming now correctly handles parallel tool calls
  • Fixed finish_reason mapping between providers

v0.4.1 — 2025-02-20 ​

Added ​

  • Multimodal support — Vision (image input) for GPT-4o, Claude, and Gemini models
  • Base64 image encoding — Send images as data URLs
  • Image URL support — Reference images by URL

Fixed ​

  • Fixed streaming interruption when upstream provider sends empty chunks
  • Fixed token counting for multi-part messages

v0.4.0 — 2025-02-01 ​

Added ​

  • Multi-model routing — Access 1400+ models through a single API
  • OpenAI-compatible API — Drop-in replacement for OpenAI's /v1/chat/completions
  • Streaming support — Full SSE streaming with all models
  • Function calling — Tool use support across providers
  • JSON mode — Structured output with response_format
  • Rate limiting — Per-key rate limits with standard headers
  • Dashboard — Usage tracking, API key management, billing

Models at Launch ​

  • GPT-4o, GPT-4o Mini
  • Claude 3.5 Sonnet, Claude 3.5 Haiku
  • DeepSeek V3, DeepSeek Chat
  • Gemini 1.5 Pro

Versioning Policy ​

SandBase follows semantic versioning for API changes:

  • Major version (v1.0, v2.0): Breaking changes to existing endpoints
  • Minor version (v0.5, v0.6): New features, new endpoints, new models
  • Patch version (v0.5.1, v0.5.2): Bug fixes, performance improvements

Breaking Change Policy ​

We avoid breaking changes whenever possible. When necessary:

  1. Breaking changes are announced at least 2 weeks in advance
  2. Deprecated features continue working for at least 30 days
  3. Breaking changes are clearly marked in the changelog
  4. Migration guides are provided for significant changes