Core Concepts

The basic concepts that make up Jinba App Neo

Jinba App Neo revolves around “agents,” with several concepts working together around them. This page explains what each one does and how they relate.

The big picture

Workspace
├── Agents (units of conversation)
│   ├── Skills (playbooks and instructions)
│   ├── Connectors (external service integrations)
│   ├── Knowledge (reference material)
│   └── Memory (what the AI learns from conversations)
├── Chats (agent-bound / ad-hoc)
├── Tasks (scheduled runs)
└── Configure (shared pages for Skills, Knowledge, and Connectors)

Agents

An agent is a conversation partner that bundles skills, connectors, and knowledge. Each agent has a model and a system prompt, and you can create multiple agents in a workspace and switch between them by purpose.

• Examples: “Sales Analyst,” “Meeting Notes Writer,” “Code Reviewer”
• Agents can be shared with members other than their creator (Share an agent)
• You can also chat without specifying an agent (an ad-hoc chat)

Throughout this documentation, we use “agent” for the product entity you create and manage, and “AI” for the runtime that actually generates responses and performs work.

Skills

A skill is a playbook or set of instructions you want the AI to reuse. Skills are written in Markdown (SKILL.md format) and carry metadata: a name, a description, and allowed tools.

• Reuse routine know-how such as “How to create an invoice” or “Translation tone guide”
• Skills can live in two places: workspace-shared (available to every agent) and agent-specific
• During a chat, the AI may detect a procedure worth turning into a skill and suggest saving it

Details: Skills

Connectors

A connector is an integration with an external service that extends what the AI can do. Connectors provide tools via MCP (Model Context Protocol), an open standard.

• Personal connectors — Connect your own accounts for Gmail, Slack, Google Calendar, Notion, and more, and let the AI act on your behalf
• jinba-flow connection — Call published workflows as tools for the AI (Call-only / full access)
• Custom MCP — Register any MCP server, such as an internal system
• Personal Jinba integration — Let the AI operate your entire Jinba workspace (listing, creating, and running flows)

Details: Connectors

Knowledge

Knowledge is the reference material the AI grounds its answers in. You can register uploaded files, URLs, and Notion pages.

• Organize sources into folders, and even auto-sync an entire folder to an agent
• Registered sources are kept in sync; they are either loaded automatically when a chat starts, or retrieved by the AI through search

Details: Knowledge

Memory

Memory is where the AI stores what it learns through chat. There are two kinds:

Type	What is stored	Who can see it
Personal memory	Your preferences, business terminology, and background	Only you (invisible to other users)
Agent memory	Procedures and caveats for that agent	Everyone who uses that agent

You can turn learning on or off, and view, edit, or delete stored memories at any time. Details: Memory

Workspaces and organizations

• A workspace is the unit that all your data — chats, agents, knowledge, and more — belongs to. Invite members to collaborate
• Workspaces belong to an organization. You can create multiple workspaces and switch between them with the switcher in the sidebar
• Members have one of two roles: Owner and Member. Administrative operations such as changing settings and managing members are restricted to owners

Configure (shared resource pages)

The “Configure” section of the sidebar contains the Skills / Knowledge / Connectors pages, where you manage workspace-shared resources. Agents pull in the resources registered here as needed.

Agent-specific resources (skills or knowledge used by only one agent) are registered from that agent’s detail page. They do not appear on the shared pages.

Models

Responses are generated by models from Anthropic’s Claude family.

Model	Characteristics
Claude Opus 4.8	Top of the line — for complex reasoning and agentic workflows
Claude Sonnet 4.6	Balanced — the default for chat and most tasks
Claude Haiku 4.5	Fastest, lowest latency — for quick lookups and summaries

• The model is fixed per chat. To switch models mid-way, start a new chat
• You can set the model per agent. Ad-hoc chats use the workspace default model