Skip to main content

Applications

An application brings knowledge, agents, members, and channels into one workspace. Its type determines how people use it: a conversation, document search, structured insights, or recommendations. An employee uses an Assistant application to keep its identity and job connected to the tools that do the work.

Choose the collections each application can read, the agents it uses, and who can access it. A customer support application can use your help documents while a finance application uses purchasing policies and invoices.

Employees use applications​

Creating an employee creates an Assistant application with a name, portrait, goal, category, and persona. Its agents, knowledge, members, and channels use the same application settings described here. One employee can contain several cooperating agents. The employee and application share an ID, while each agent has its own ID. Existing Assistant applications keep their current behavior and appear as employees only when they have an employee profile.

In Applications, its card shows the employee portrait and an Employee link back to preparation. Open Work together from the employee to use this application’s assistant, or Advanced workspace to inspect its agents.

Employee applications show a portrait and an Employee link. Other application types keep their existing controls.
Employee applications show a portrait and an Employee link. Other application types keep their existing controls.

To add an existing Assistant application to the team, open its card menu and select Add to Employees. Review the goal and portrait, and choose the primary agent when there is more than one. This preserves the application and its published state; it adds no duplicate resources. See Bring an existing assistant onto your team.

What's inside an application​

  • Knowledge: the collections it can read. An application only sees what you attach to it.
  • Agents: one or more agents that decide how it answers and what it does.
  • Members: who can use it, and who can change it.
  • Channels: where it meets users, from a web widget to WhatsApp.
  • Settings: instructions, persona and tone of voice, environments, integrations, and more.

It scopes your knowledge​

This is the part that matters: an application reads only the collections you attach to it, nothing else.

That means a customer-facing assistant can answer from your public help docs while staying blind to internal HR files in the same company. You decide what each application knows.

You set this when you create the application, under Select Knowledge, and you can change it any time from the application's Knowledge settings. Docana also creates one collection automatically for each application, named after it, as a home for documents that belong to that application alone.

Each enabled collection also has a File Sharing toggle. It controls whether the assistant may send that collection's documents to end users as files. It's off by default, so people can ask questions about the content but never receive the underlying file until you turn it on. The application's own collection is the exception: its documents are always shareable.

Both toggles are available outside the UI too, in the REST API, the CLI (docana apps collections), and the Docana MCP (linkApplicationCollection, unlinkApplicationCollection, and setApplicationCollectionFileSharing).

Enable the collections the application can search. File Sharing is a separate setting for sending source files to users.
Enable the collections the application can search. File Sharing is a separate setting for sending source files to users.

Choose a type​

When you create an application, you pick a type. The type decides how people interact with it.

Choose an application type, enter its name, and write the instructions. This example creates a Documentation Assistant.
Choose an application type, enter its name, and write the instructions. This example creates a Documentation Assistant.

Assistant​

A chat interface backed by your knowledge. People ask questions in plain language and your agents answer from your documents, with sources. This is the most common type, and the right one for support bots, internal help desks, and any "ask us anything" experience. Deploy it to a web widget, WhatsApp, Slack, or Teams.

Document Insights​

This type appears as Documents Insights in the application picker.

Turn a stack of documents into a structured table. You define the columns you want, point it at your documents, and it extracts every field into a grid you can sort, filter, and query. Right for comparing many documents at once, like pulling renewal dates from every contract. See Document Insights.

A search interface over your documents. People type what they're looking for and get ranked results with sources, without a back-and-forth conversation. Right when discovery matters more than dialogue, like an internal knowledge portal. See Enterprise Search.

Recommendations​

Beta

Recommendations is in beta.

Suggests related items from your knowledge: similar contracts, related documents, or products that fit. Right for "you might also want" experiences built on your own data.

The Assistant opens with a message composer and shortcuts to agents, knowledge, uploads, and settings.
The Assistant opens with a message composer and shortcuts to agents, knowledge, uploads, and settings.

Agents work as a team​

An application can hold more than one agent, and they can work together. Instead of building one giant agent that does everything, you build a few focused ones and let them cooperate, the way a team splits up work.

There are two ways to connect them:

  • Sub-agents: one agent hands a task to another and uses the result. Your main agent can call a "contract specialist" agent for the legal parts and a "pricing" agent for the numbers, then combine both. Each sub-agent stays focused on what it's good at.
  • Coordinator: mark an agent as a Coordinator and it waits for the other agents to finish, then receives their results as context and makes the final call. It's the team lead that reads everyone's work and writes the summary.

Smaller, focused agents are easier to test and improve than one that tries to do everything. See Creating an Agent and Testing Agents.

Persona and tone of voice​

By default, an application speaks in Docana's standard voice. The Persona page in its Settings changes that. Pick a persona, and the assistant, the agents, and every channel (web, widget, WhatsApp, Slack, Teams, and voice calls) answer in that voice.

The persona gallery on the application's Persona settings page, with voice styles and use-case presets
The persona gallery: voice styles on top, presets tuned for common jobs below

You start from a preset. Some are voice styles, like Professional, Empathetic, or Storyteller. Others are tuned for a job, like Customer Support, IT Helpdesk, or Legal. Neutral is the default, and keeps the voice exactly as it is today.

From there you can fine tune the voice with six sliders (formality, warmth, humor, energy, detail, and emoji), add voice notes in your own words, such as "always greet returning customers by name", and preview a sample reply before saving.

Persona fine-tuning with trait sliders, voice notes, and a sample reply preview
Fine tune the voice with sliders and notes, and preview a sample reply

A persona changes how the application communicates, not what it knows or does. Your instructions still define the behavior, and a persona never overrides them. For the full walkthrough, see Creating an Application.

Like the other application settings, the persona is also available outside the UI, in the REST API, the CLI (docana apps persona), and the Docana MCP (the updateApplication tool), so agents can configure the voice too. The API reference documents every preset and slider.

Create one​

  1. Go to Applications and click + Create Application
  2. Pick a type
  3. Give it a name, choose a group, and select the knowledge it can read
  4. Add members if you want, then click Create

For the full walkthrough, see Creating an Application.

Next steps​

Organize employees and applications​

The Category field uses the same company catalog in both creation forms. Administrators can type a new name, select Create, and choose an icon in place. Company → Categories manages names and icons for everyone in that company. Renaming a category keeps its existing applications and employees together. The catalog includes 20 industry defaults and supports company-specific categories.

Export and import a setup​

Open an application card's menu and select Export application to download its definitions and a list of source connections. Import application above the application list restores a package as a new private draft. The same format supports employees, including their portrait and goal.

The package carries draft agents, evaluation definitions, routines, insights configuration, mock baselines, assistant instructions, application-local persona and capability settings, document templates, and recommendation definitions. It carries metadata references for knowledge and external connections. It does not copy binaries, credentials, memberships, conversations, execution history, or published revisions. Company defaults and external resources need review in the destination.

After import, use Import checklist in the application card menu to inspect the original resources, reconnect them, and update workflow references. Agents are unpublished and schedules are paused until you review and enable them. This is a definition package, not a database backup.