Platform Overview¶
Manifest Platform is an enterprise operating system for orchestrating data, agents, and workflows. It provides the infrastructure, security, and governance layer that organizations need to build, deploy, and manage AI-powered applications at scale.
Core Concepts¶
Organizations and Workspaces¶
The platform uses a hierarchical tenant model:
- Organization -- the top-level tenant, representing a company or business unit. All resources, users, billing, and configuration are scoped to an organization.
- Workspace -- a logical grouping within an organization for separating projects, teams, or product lines. Each workspace has its own components, services, and deployments.
- Environment -- a deployment target within a workspace (e.g.,
dev,stage,prod). Environments control where code runs and which infrastructure credentials are used.
graph TD
Org[Organization] --> WS1[Workspace: Product A]
Org --> WS2[Workspace: Product B]
WS1 --> E1[Environment: Dev]
WS1 --> E2[Environment: Stage]
WS1 --> E3[Environment: Prod]
WS2 --> E4[Environment: Dev]
WS2 --> E5[Environment: Prod]
See Organizations for details.
Solutions and Components¶
Everything you build on the platform is a component -- a versioned, deployable unit of functionality. Components include:
| Component Type | Purpose |
|---|---|
| Agents | AI agents with tool access, memory, and structured outputs |
| Workflows | Multi-step automation pipelines |
| Code Blocks | Reusable Python functions |
| Tools | Capabilities exposed to agents |
| Models | LLM model configurations and provider settings |
| Prompts | Versioned prompt templates |
| Schemas | Data structure definitions |
| Transforms | Data transformation logic |
| Hosted Services | Backend API services with managed endpoints |
| Connectors | Integrations with external systems and databases |
Components are grouped into Solutions -- packages that bundle related components for deployment and sharing.
Hosted Services¶
Hosted Services let you deploy backend logic as managed API endpoints. The platform handles routing, auth, versioning, and scaling. Endpoints can be exposed as agent tools through the Skills System.
AI Gateway¶
The AI Gateway provides a unified interface to multiple LLM providers. All AI model calls flow through the gateway, which handles routing, authentication, rate limiting, cost tracking, and fallback logic.
Connectors¶
Connectors integrate the platform with external systems -- databases, SaaS APIs, cloud storage, and more. Each connector type defines a schema for credentials and configuration. Connector instances are scoped to workspaces and can be bound to hosted services and workflows.
Deployment Rings¶
Deployment rings provide graduated rollout for components. A typical setup includes:
- Ring 0 (Canary) -- small percentage of traffic for early validation
- Ring 1 (Early adopters) -- broader rollout
- Ring 2 (General availability) -- full production traffic
Rings enable safe promotion of changes from development through production.
How Components Fit Together¶
graph LR
User([User / Agent]) --> GW[Service Gateway]
GW --> HS[Hosted Service]
HS --> CB[Code Block]
CB --> AIGw[AI Gateway]
CB --> Conn[Connector]
CB --> DS[Dataset]
AIGw --> LLM[LLM Provider]
Conn --> Ext[External System]
Agent([AI Agent]) --> Tools[Tools / Skills]
Tools --> GW
Agent --> AIGw
A typical flow:
- A user or agent makes a request.
- The Service Gateway routes it to the correct Hosted Service endpoint.
- The endpoint's Code Block executes, potentially calling the AI Gateway for LLM inference or a Connector for external data.
- AI Agents discover available endpoints through Skills and invoke them as tools.
Platform Capabilities¶
Security and Access Control¶
- Role-based access control (RBAC) with organization, workspace, and resource-level permissions
- Scope-based authorization on every API endpoint
- Audit logging for all management and runtime operations
- Multi-factor authentication support
- API key management with expiration and revocation
Observability¶
- Request tracing across service invocations
- Gateway trace logging with configurable retention
- Credit consumption tracking with attribution to users and resources
- Alert system for balance, burn rate, and anomaly detection
Billing¶
The platform uses a credit-based billing system where all resource consumption (LLM calls, compute, storage, data transfer) is metered and charged against an organization's credit balance. Rate cards define the pricing for each service type.
Developer Experience¶
- Flow SDK -- Python SDK for building and deploying components
- CLI -- Command-line interface for deployment, testing, and management
- Auto-generated API docs -- Swagger UI for every hosted service
- Client stub generation -- Python and TypeScript clients generated from endpoint schemas
- Draft preview -- Test endpoints before publishing
API Base URL¶
All platform API requests go through the gateway:
Tenant-scoped resources follow the pattern:
Runtime service invocations use the service gateway: