# Chat and agents

Your pools reach AI assistants in two ways: files you add to a chat land in your AI Chat pool, and agents query your other pools through the Data Pools MCP server.

# The AI Chat pool

Every user has an AI Chat pool, created automatically. Files you upload in PHOENIQS Chat for file search land in it, and get the same treatment as any upload: OCR, indexing, a description and metadata.

It is yours alone. Only you see it in Data Pools.
Chat already uses it. The assistant searches these files through Chat's own file search, so the pool is deliberately never offered to agents through the MCP server. That would give them the same documents twice.
Files attached to a shared agent stay searchable for everyone who may use that agent, but only through the agent.
It cannot be deleted or renamed. You can delete individual documents from it.

# Connecting agents to your pools

Agents connect to Data Pools through one read-only MCP server. MCP Access in the Data Pools menu shows its address, called the MCP endpoint.

On first use, the agent asks you to sign in once with your PHOENIQS account. From then on it sees exactly the pools you can read in Data Pools: your own, those you are a member of, and those open to your organisation. The AI Chat pool is the exception, as described above.

Uploads stay in the app

Through the MCP server, agents can only list, search and read. Creating pools, uploading and deleting happen in the Data Pools app.

# PHOENIQS Chat

When you provision PHOENIQS Chat, you can connect it to one of your company's Data Pool Dedicated. The chat then gets the Data Pools MCP server, stores its File Search files in Data Pools, and shows a link to Data Pools in its sidebar.

A chat connects to one Data Pool at most. It is chosen when the chat is provisioned and cannot be changed afterwards.
One Data Pool can serve several chats.
While a chat is connected to it, the Data Pool cannot be cancelled.

# OpenCode

OpenCode on your own computer connects with your own sign-in. Add the server to your opencode.json, with the MCP endpoint from MCP Access:

{
  "mcp": {
    "datapool": {
      "type": "remote",
      "url": "https://<your Data Pools address>/api/mcp"
    }
  }
}

Then sign in once:

opencode mcp auth datapool

Your browser opens for the PHOENIQS sign-in. There is no client ID or secret to set up.

# The tools agents get

Tool	Pools	What it does
`list_data_pools`	All	Lists the pools the user can read, with each pool's description and a summary of every document or table in it
`search`	Knowledge, SharePoint	Searches one or more pools by meaning and keyword. Can be narrowed to a person, organisation, topic, location or date range.
`filter_content`	All	Finds documents and tables by their metadata, such as "every document that mentions this company"
`list_metadata_facets`	All	Lists the people, organisations, topics and locations found in the pools, with how often each appears
`list_documents`	Knowledge, SharePoint	Lists a pool's documents, with their descriptions, metadata and OCR confidence
`get_document_content`	Knowledge, SharePoint	Returns a document's text. For a SharePoint file that is only listed, its link.
`get_document_structure`	Knowledge, SharePoint	Returns the fields, tables, line items and totals read from a scan or invoice
`list_tables`	Database, External	Lists a pool's tables, with their columns, types and descriptions
`query_table`	Database, External	Returns rows, with filters such as equals, greater than or contains, sorting and paging
`aggregate_table`	Database, External	Counts, sums, averages, minimums and maximums, optionally grouped by up to three columns

# What agents see of a pool

A pool's Details section shows it from the agent's side:

Row	What it shows
Pool ID (for MCP tools)	The ID agents use to name this pool
Pool type	Knowledge, Database, External database or SharePoint
Embedding model	The model that indexes the pool for search by meaning
Access	Private or Organisation. See Access.
MCP exposure	Exactly what `list_data_pools` returns for this pool: your description, the AI summary of its contents and every document or table with its description
Applicable MCP tools	The tools that work on this type of pool

The pool's description is the first thing an agent reads. It is set when you create the pool and cannot be changed in the app afterwards, so write it with agents in mind: what the pool holds, and what questions it can answer.

Access

../access/

PHOENIQS Chat

../../phoenixchat/

data pools mcp agents phoeniqs chat opencode tools