Set up
Knowledge
Knowledge is what agents can look up beyond the conversation: workspace documents everyone shares, and an optional memory provider that agents search and save to.
- Documents are handbooks, policies, and notes that owners and admins write in Settings → Knowledge. Every agent in the workspace can search them.
- A knowledge provider (Mem0, Supermemory, Zep, or Cognee) is an external memory an agent searches and, with most providers, saves to. Each agent connects to at most one.
Neither includes conversations. Past chats are only searchable by the person who had them; see Chats and tool activity.
Add a document
- Open Settings → Knowledge and choose Add document.
- Give it a title and write or paste the content, in plain text or Markdown.
- Choose Create document.
- Title: 1 to 200 characters.
- Content: up to 50,000 characters.
Owners and admins can add, edit, and delete documents. Members can read them but not change them. Deleting a document can't be undone, and agents can no longer read it.
How agents use documents
Agents search documents by keyword, then read the ones that match. Search covers each document's title and content, using PostgreSQL full-text search with English word matching. Agents are told to cite the titles and IDs of the documents they use.
Document tools are part of an agent's Harness tools, which also cover plans, skills, and history search. An agent with Harness tools turned off can't read documents.
Agents can also create or update documents from a chat. Like edits on the Knowledge page, saves need permission to edit documents, so they work only in chats with owners and admins, and they're recorded in the activity log.
Connect a knowledge provider
Owners and admins add knowledge connections in Connections, under Knowledge; setup offers one as the optional memory step. The connection holds the endpoint and credentials. What to search, and how many results to bring back, is set per agent.
Choose the knowledge provider for new agents in Settings → Defaults → Knowledge, and for one agent in its Settings tab under Knowledge. Settings → Defaults → Knowledge also controls availability for the workspace:
- Off for everyone: no agent can use a knowledge provider. Documents stay available.
- Off by default: off unless turned on for an agent.
- On by default: copied into new agents. Existing agents keep their settings.
A knowledge connection is shared by the whole workspace, so anything its credentials can reach is reachable by every agent that uses it. Scope those credentials to what everyone in the workspace may see.
How agents use a knowledge provider
An agent with a knowledge provider gets three tools:
- Search knowledge: search by keywords. Agents are told to cite the titles, IDs, and source URLs returned.
- Read document: read one result by its ID.
- Save knowledge: add something new. All four providers support it.
Saves are append-only: they add a memory or document and can't replace an existing one. Each provider processes what it receives in the background, so a save returns a receipt rather than the saved item. The new memory shows up in search once the provider has processed it.
Provider settings
Each provider needs an API key on the connection. The other settings are per agent; unset settings use the provider's own defaults.
Agents that use the same scope (agent ID, container tag, user or graph ID, or dataset) share memories. Give agents different scopes to keep their memories apart.
Mem0
- Mem0 agent ID: the scope this agent's memories live under. Defaults to
adapt-agent. - Result limit: unset uses Mem0's default of 10.
- Minimum relevance: unset uses Mem0's default of 0.1.
- Rerank results and Include expired memories:
trueorfalse. Unset uses Mem0's default,false.
Supermemory
- Container tag: the scope for this agent's memories. Defaults to
adapt-agent. - Search result limit: unset uses Supermemory's default of 10.
- Minimum relevance: unset uses Supermemory's default of 0.5.
Zep
- Graph type:
userorgraph. Defaults touser. - Zep user or graph ID: defaults to
adapt-agent. - Search result limit: unset uses Zep's default of 10.
Cognee
Cognee's connection also needs a Base URL: your Cognee Cloud tenant, such as https://your-tenant.aws.cognee.ai, or a self-hosted Cognee server the Adapt server can reach. The server must use API-key authentication with backend access control on.
- Dataset name: defaults to
default_dataset. - Result limit: unset uses Cognee's default of 15.