Define Work
Goals, Inputs, Instructions, Expected Output & Success Criteria.
TaskInputsOutput
SANDBASE · Build / Configure / Deliver
Turn a task into a deliverable Agent.
Define goals, configure Models / APIs / Skills, test once in a live run, and choose to deliver to a product, AI tool, or scheduled task, all within a continuous workspace.
No longer switching between three pages
1Continuous Agent Builder
3Delivery Method
1Unified Execution Trajectory
01 · Assemble AGENT
Define work, configure capabilities, test execution, and then deliver.
Each step continues on the same Agent Configuration, no longer splitting work definition, capability configuration, and release into three mutually repetitive products.
Configuration Capability
Select a model, connect to real services, and set permissions for each action.
Run Validation
Check Results, Call Stack, Duration, Cost, and Fail Recovery.
ResultsTrackCost
DeliveryAPI EndpointUsing in AI ToolsJob Scheduling
02 · Service Architecture
Two entry points, one Agent execution system.
After workflow definition, Agent and AI tools connect via MCP; products, backend, and self-hosted Agents connect via API/SDK. Both paths share authorization, policies, execution, and auditing.
API / SDK / MCP
AuthPolicyRuntimeAudit
sandbase API／SDK／MCP
Gmail
Slack
Notion
HubSpot
GitHub
Stripe
Google Docs
Webflow… AND 100+ MORE SERVICES03 · AGENT BUILDER
Define work as executable configuration.
Left side defines the goal, middle configures models and services, right side clarifies results and permissions. Customers see not a concept diagram, but an Agent ready to run.
Identify accounts with renewal risk and notify the correct owner.
Identify risk · update CRM · send one actionable alert
Claude SonnetReasoning + structured output
Stripe→HubSpot→Slack- StripeRead Subscription
- HubSpotUpdate Risk Status
- SlackSend alert
✓ User-scoped authentication
04 · Test Run
Prove it can run before delivery.
Verify results, service call traces, latency, cost, and error recovery in a single test, turning 'it works' into verifiable evidence.
01StripeSubscription Risk Detected1.2s
02HubSpotRisk status has been updated2.1s
03SlackResponsible person has received a reminder0.8s
05 · Choose delivery method
With the same Agent, you can choose the most suitable delivery method.
After testing, there's no need to rebuild. Bring the same configuration into the AI tool, publish as a product Endpoint, or hand it over to a scheduled task for continuous operation.
Using in AI Tools
Let Claude, ChatGPT, Codex directly call this Agent via MCP.
MCP · ONE CONNECTIONView AI Tools Integration →API Endpoint
Access SaaS, Backend, or your own Agent via API/SDK.
API / SDK · VERSIONEDView Developer Access →Job Scheduling
Trigger by time, event, or business status, and retain complete execution history.
TRIGGER · MONITORSet Schedule →06 · Getting Started with AGENTS
Start with executable work, not an empty configuration.
Choose a mature workflow, modify the goal, capabilities, and delivery method, then turn it into your own Agent with one test.
Sales Lead Research Agent
Sales Lead Research Agent
Exa→HubSpot→NotionResearch new leads, supplement company and contact information, and generate a complete research record.
- Inputs
- New Lead
- Delivery
- CRM + Research Notes
Customer Risk Agent
Renewal Risk AgentStripe→HubSpot→SlackIdentify payment or renewal anomalies, update customer risk status and notify the team.
- Inputs
- Payment Status Change
- Delivery
- Risk Records + Alerts
Engineering Response Agent
Engineering Incident Response Agent
Sentry→GitHub→
LinearRead high priority errors, locate related code, and create engineering tasks.
- Inputs
- High Priority Errors
- Delivery
- Diagnose + Engineering Tasks
Daily Summary Agent
Daily Summary AgentGmail→Notion→SlackOrganize important emails and to-dos, generate a summary ready to deliver to your team.
- Inputs
- Daily Business Insights
- Delivery
- Summary + Notification
Content Publishing Agent
Content Publishing AgentGoogle Docs→Webflow→SlackReads approved content, publishes pages, and notifies the team with the final link.
- Inputs
- Approved Documents
- Delivery
- Page + Link
Start from Your Own Work
Define input, result, and delivery locationJob
Capabilities
Success Metrics
Delivery Channels
Three-step start
Build and deliver without switching between three pages.
Define work and configure capabilities.
Write clear goals and results, while adding Models, APIs, Skills, and permissions.
BUILD + EQUIPTest with a real run.
Check results, traces, duration, and cost until success criteria are met.
TEST + VERIFYSelect Delivery Method
Publish as Endpoint, Bring in AI Tools, or Set Up Scheduled Tasks.
DEPLOY02 · Connect and Build→Next Step
Agent is configured, let it deliver at real pace now.
Set triggers, time zones, retries, and delivery goals, or check a live run directly.