SandBase is an Agent-native Platform, positioned as a platform for production AI agents, claiming to provide sandboxes, model routing, MCP servers, skills, sessions, and a unified API (this description comes from the product's English introduction; the official website body does not elaborate on the corresponding details). Its core goal is to let users find models, APIs, skills, and agents, and turn them into traceable workflows to complete real work.
I. Main Capabilities The platform brings together 1,195 models and 1,105 real-world APIs, and each capability card displays the price and includes a runnable request example. The Store provides reusable agent capabilities, including models, APIs, ready-made agents, and skills, which users can test, copy, install to Setup, or clone to Build Agent. The platform also provides the following example capabilities:
- AGENT SEO Keyword Insights: Use DataForSEO to discover, validate, and prioritize keywords, run by usage.
- LLM openai/gpt-6.1-sol: The DevDay 2026 upgrade of OpenAI GPT-6 Sol, 1 million token context window, $2.00 input · $10.00 output / 1M.
- IMAGE google/nano-banana-2.1/edit: Google's fast image editing, can blend up to 14 reference images, $0.0357 / time.
- API exa/search: Exa neural web search, returns page content inline, free / time.
- LLM Claude Opus 5.5: Anthropic's strongest Claude 5 model, used for complex reasoning and coding, $3.80 input · $19.00 output / 1M.
- AGENT YouTube Transcript: Extract transcripts and subtitles for research and summarization, free to clone.
- API Instagram: Get profiles, posts, Reels, and comments for creator research, see endpoint for price.
- SKILL YouTube Research: Research videos, channels, and audiences from AI tools, public skill.
II. How to Use The platform provides three ways to put capabilities to work:
- Setup: Use in local AI tools. From SandBase select useful capabilities, run Setup once, and you can use them in Codex, Claude, Cursor, or Kiro. The setup address is https://www.sandbase.ai/SKILL.md（可复制 setup prompt).
- API: Call APIs. Copy available requests for models, search, scraping, data, or SaaS APIs, with prices shown on the cards.
- Agent: Build agents. Define tasks, assign tools, test, then publish or schedule runs. One key covers all three ways above.
III. Workflow Building Users can define tasks and give Agents appropriate capabilities, see one run result first before letting anything depend on it, then deliver it to AI tools, APIs, or schedulers. Each execution becomes a run, with inputs, outputs, traces, and status, and Activities shows the cost. The platform also supports notifying systems via webhook when asynchronous models, APIs, or agent work completes or fails, without polling.
IV. Applicable Scenarios and Audience The platform is aimed at builders who need to quickly obtain usable AI capabilities, and FDEs who need to turn customer requirements into reusable agent workflows. Agent workflows can be used for repeatable processes such as researching customers, enriching leads, monitoring competitors, and classifying GitHub issues. The example shows a process that turns a business problem into a customer brief with source citations.
V. Billing
- Start: $0, use introductory credits to explore models, APIs, and agent workflows.
- Build: Pay-as-you-go, pay for the underlying models, APIs, and runtime work used by calls.
- Scale: Custom, centralized workspace, control, and support for production teams. The platform claims savings of up to 25% on models, free APIs, and that the displayed price is the price paid.
VI. Limitations Specific technical details about sandboxes, model routing, MCP servers, sessions, etc. are not yet provided in the main text of the official website; sandboxes, model routing, MCP servers, skills, sessions, etc. appear only in the product introduction and lack further explanation. Some API pricing (such as Instagram) requires checking the endpoint, and specific numbers are not yet given.
