Glossary.
Key terms used throughout The Intern platform and documentation.
Agent
An AI operative deployed to handle a specific business role — sales closer, receptionist, researcher, support hero, etc. Each agent has its own identity, system prompt, skills, memory, and credit balance. Agents are stored in the user_agents table in the database.
Agent Template
A pre-built agent configuration stored in the agent_templates table. Templates include a name, role, system prompt, skill list, tool list, and estimated ROI. Users clone templates to create their own agents via the /api/templates endpoint.
Skill
A modular capability installed into an agent. Skills are defined in SKILL.md format and stored in the skills table. Each skill has a category, trigger conditions, and a set of actions. Skills can be built-in (isBuiltin: true) or custom.
Workspace
The persistent context layer for a user. Workspace files are markdown documents (CONSTITUTION.md, IDENTITY.md, NOW.md, THREADS.md, RECENT.md) stored in the workspace_files table. The agent reads these on every conversation to maintain continuity.
Brain-Graph (Memgraph)
The long-term structured memory layer. Replaces simple vector embeddings with a rich knowledge graph. Extracted facts are stored as nodes and relationships, allowing the agent to perform complex reasoning (e.g. tracking dependents, caller graphs, and organizational structure).
Canvas Workflow UI
The node-based visual editor where you design agent execution paths. Instead of a linear chat, the Canvas allows you to define states, decision branches, and tool execution nodes that the agent treats as a state machine.
Credits
The billing unit. 100 credits = $1 USD. Credits are consumed by AI model calls, tool executions, and API requests. Balances are stored per-user in the users table. Transactions are logged in the credit_transactions table.
Orchestrator
The multi-agent routing layer at /api/orchestrator. Uses LLM intent classification to route incoming requests to the most capable agent. Enforces RBAC so agents only access data they are permitted to see.
Alpha Engine
The advanced reasoning layer at /api/alpha. Powers sales negotiation, legal analysis, competitive strategy, and financial modeling using Claude with specialized system prompts.
Inbox
The inter-agent messaging system. Agents send structured messages to each other via the inbox_messages table. Messages have a priority (low/normal/high/urgent) and status (pending/read/actioned/archived).
Pending Decision
A high-risk action that requires human approval before execution. Stored in the pending_decisions table with a decision type, context, and status (pending/approved/rejected).
BYOK
Bring Your Own Key. Users can store their own Anthropic API key in Settings → API Keys. It is saved in the users.preferences JSON column and used in place of the platform key for all their agent calls.
Durable Object
A Cloudflare Workers primitive used for stateful agent sessions. The AgentSession Durable Object maintains real-time conversation state across streaming requests.