What is ScaiCore?

Building AI agents today typically involves cobbling together prompts, API calls, tool integrations, and orchestration logic across multiple languages, frameworks, and services. It works, but it's fragile, hard to test, and difficult to reason about.

ScaiCore takes a different approach. It's a dedicated programming language where everything an agent needs — its personality, constraints, tools, memory, workflows, and collaboration rules — is defined in one place, in one syntax, with one deployment model.

A ScaiCore agent is called a Core. Cores range from simple assistants that answer questions using a knowledge base to complex multi-agent teams that orchestrate entire business processes with human oversight at critical decision points.

The Languagescaicore

@core CustomerSupport {
    @identity {
        name = "Support Assistant"
        version = "1.2.0"
        description = "Handles customer inquiries with empathy and precision"
    }

    @personality {
        tone = "professional, warm"
        verbosity = "concise"
        languages = ["en", "nl", "de"]
    }

    @constraints {
        never "Make commitments about refunds without authorization"
        never "Share other customers' information"
        always "Verify customer identity before sharing account details"
        prefer "Offer alternative solutions when primary solution unavailable"
    }

    @plugins {
        knowledge = scailabs/knowledge-base@1.0
        crm = company/salesforce-connector@1.0
    }

    @llm {
        primary = scaigrid { model = "claude-sonnet-4-20250514" }
    }
}

Key Concepts

Strict

The Core follows predefined paths with no deviation. Fully deterministic.

Guided

The Core can make decisions within explicitly defined boundaries.

Adaptive

The Core reasons freely but respects its constraints.

Autonomous

The Core has maximum freedom, with constraints as guardrails.

The Console

Deployment management

Deploy, scale, and version Cores through a visual interface

Execution monitoring

Real-time and historical visibility into Core execution

Approval center

Unified inbox for human-in-the-loop checkpoints across all Cores

Debugging tools

Trace execution flows, inspect memory state, review decision paths

Analytics

Execution metrics, LLM costs, performance trends

Registry browser

Discover and deploy Cores from a shared registry

Use Cases

Customer support agent

A Core that handles inquiries, checks the knowledge base, creates tickets, and escalates to humans when it encounters situations outside its confidence level.

Document processing pipeline

A team of Cores that ingests documents from ScaiDrive, extracts structured data, validates it against business rules, and routes the results — with human checkpoints for ambiguous cases.

Workflow automation

Cores that monitor events (new emails, file uploads, system alerts), apply business logic, and take action — with the rigidity spectrum controlling how much independent judgment they exercise.

Compliance monitoring

An agent that reviews content, communications, or transactions against regulatory rules, flagging potential issues for human review.

Learn more

Read the Language Reference →