SandBase

What is SandBase

SandBase is an agent-native delivery platform where you can find models, APIs, skills, and agents, then assemble them into traceable workflows that actually get work done. Instead of manually stitching together separate services, SandBase provides a unified layer: browse a capability store, define a workflow, test it, and deliver it to an application, an AI tool, or a schedule. The platform supports three main modes. First, you can use capabilities directly inside local AI tools — pick useful models, APIs, or agents, run a one-time setup, and they appear in Codex, Claude Code, Cursor, or Kiro. Second, you can call any individual capability through an API by copying a working request for a model, search, scraping, data, or SaaS endpoint. Third, you can build a reusable agent: define the job, give it tools, test the workflow, then publish or schedule it. A concrete example from the site shows a customer research question processed by a workflow that searches live sources, adds company data, and builds a source-backed brief, delivered as a customer brief.

Key features

• Capability store: Browse and select from Models (reasoning, vision, audio, image, and embedding), Search APIs, Data extraction, SaaS APIs, Agent templates, and Skills.
• Local AI tool setup: A single connection via a curl script (curl -fsSL https://sandbase.ai/install.sh | sh -s -- --client codex) makes selected capabilities available inside supported tools: Codex, Claude Code, Cursor, Kiro, Windsurf, OpenCode, Gemini CLI, Qwen Code, Warp, ChatGPT, Claude Desktop, and Kimi CLI. The setup uses guided authorization and avoids putting keys in shell history.
• Direct API access: Copy a working request for any capability and receive a structured response in your product. This includes search, scraping, data extraction, and SaaS integrations.
• Agent builder: Define the job, assign tools, test the result, then publish or schedule the agent. Start from working agent templates and adapt them to your use case.
• Traceable workflows: From business question to deliverable, every step is visible and reusable, so a research workflow can be handed over as a deliverable asset.
• Delivery options: Deliver the finished workflow to an application, an AI tool, or a schedule.
• Skills: Reuse packaged instructions and code inside agent workflows.

How to use it

1. Browse the SandBase store and choose the models, APIs, skills, or agents you need.
2. For local AI tools, run the setup script with the --client flag for your tool (e.g., codex). Authorize once using the guided flow.
3. Select the capabilities you want that tool to access. They become available as tools inside the AI tool.
4. To call a single capability from your product, copy the API request from SandBase and integrate it into your app. A request travels from your product to the capability, and a structured response returns.
5. To build a reusable agent, define the job, give it the right tools, test the workflow, then publish or schedule it.
6. Deliver the finished workflow to an application, an AI tool, or a schedule where the work happens.

Who it is for

SandBase is aimed at builders and Forward Deployed Engineers (FDEs) who need to connect AI agents to real-world systems. If you already use local AI tools like Claude Code, Cursor, or Codex and want to add live web search, data extraction, or SaaS integrations without managing separate API keys, SandBase Setup simplifies that. Teams that need repeatable, source-backed research or customer briefs can benefit from reusable agent workflows. It is also useful for developers who want to call a single model, search, or scraping API without building a full integration.

What to know before you use it

SandBase does not execute the actual models, APIs, or skills itself; it orchestrates third-party capabilities, so availability, rate limits, and output quality depend on those external services. The setup script requires command-line access and some technical comfort, which may not suit non-developers. Workflow traceability is provided, but you should verify outputs from agents against authoritative sources when using them for critical decisions. Pricing details are not listed on the source page, so check the official site for current plans. If your use case requires guaranteed uptime, private data handling, or compliance with strict data residency, evaluate whether a direct, first-party API integration is more appropriate than routing through SandBase.