Agent Knowledge Bases: Built-In RAG Retrieval Pipeline

Knowledge Bases give your agents access to company data — Jira issues, Confluence pages, GitHub repositories, Notion workspaces, SharePoint libraries, Google Drive files, Salesforce records, and more — without hardcoding any of those sources into the prompt. The full RAG pipeline (chunking, embedding, hybrid search, and reranking) runs entirely inside Archestra. No external vector database or separate retrieval service is required. When at least one knowledge source is assigned to an agent, Archestra automatically adds the built-in query_knowledge_sources tool. The model calls it during a run to search across all assigned sources and pull the most relevant documents into its answer.

Configuration

Before creating a Knowledge Base, open Settings > Knowledge and configure both an embedding model and a reranking model. Neither Knowledge Bases nor connectors can be used until both are set.

Embedding Configuration

Pick the API key and embedding model. The model vectorizes ingested documents for semantic search. It is locked once saved — both indexing and querying must use the same model. To change it, click Drop to clear the index; all documents are re-embedded on the next sync.Supported embedding dimensions: 768, 1536, 3072. If your key does not appear, go to LLM Providers > Models, sync the provider, and set the dimensions for the embedding model.

Reranking Configuration

Pick the LLM key and chat model that scores and reorders search results by relevance. Any configured LLM provider key and chat model can be used here — this is separate from the embedding model.

How Retrieval Works

Connectors run on a cron schedule, pulling documents that are chunked and embedded into PostgreSQL with pgvector. At query time, the agent’s input is embedded and then both vector search and optional full-text search run in parallel. Results are fused using Reciprocal Rank Fusion, reranked by the configured LLM, and filtered by ACL before being returned to the agent.

The entire RAG stack runs within PostgreSQL — no external vector database is required.

Creating a Knowledge Base

A Knowledge Base is a named collection of connectors. Create one from the Knowledge page, then assign one or more connectors to it. The same Knowledge Base can be reused across multiple agents and MCP Gateways.

Open the Knowledge page

Navigate to Knowledge in the left sidebar and click New Knowledge Base.

Name and save

Give the Knowledge Base a descriptive name and click Create.

Add connectors

Inside the Knowledge Base, add one or more connectors to pull data from your external systems.

Assign to an agent

Open the agent’s edit dialog, scroll to Knowledge Sources, and select the Knowledge Base.

Files (Static Uploads)

Files are static documents uploaded from Knowledge > Files and assigned directly to agents or MCP Gateways. Use files when you want reusable retrieval from individual documents without setting up an external connector. Supported file types: .txt, .md, .csv, .json, .xml, .pdf Files use the same visibility model as other knowledge resources:

Mode	Behavior
Owner	Only the uploader can view and query the file.
Teams	Only members of selected teams can view and query the file.
Organization	Anyone in the organization can view and query the file.

Chat attachments stay with one conversation. To reuse an attachment across sessions, save it to Knowledge > Files from within the chat.

Connector Visibility

Each connector has a visibility setting that controls which users can retrieve its data when an agent calls query_knowledge_sources. Users only see sources they have access to, and only those can be assigned to agents.

Mode	Behavior
Org-wide	All documents accessible to every user in the organization.
Team-scoped	Documents accessible only to members of the assigned teams.
Auto-sync permissions	ACL entries synced from the source system (user emails, groups). Coming soon.

Users with the knowledgeSource:admin role can view and query every connector regardless of visibility.

Team-scoped visibility and auto-synced ACLs require an enterprise license. Contact sales@archestra.ai for licensing information.

Supported Connectors

Jira — Issues and Comments

Syncs issue descriptions, comments, and metadata from Jira Cloud or Server.Authentication: Atlassian account email and an API token.

Field	Description
Base URL	Your Jira instance URL (e.g., `https://your-domain.atlassian.net`)
Cloud Instance	Toggle on for Jira Cloud, off for Jira Server/Data Center
Project Keys	Comma-separated project keys to include (optional)
JQL Query	Custom JQL to filter issues (optional)
Comment Email Blacklist	Comma-separated emails whose comments are excluded (optional)
Labels to Skip	Comma-separated issue labels to exclude (optional)

Confluence — Wiki Pages

Syncs pages from Confluence Cloud or Server.Authentication: Same Atlassian email and API token used for Jira.

Field	Description
URL	Your Confluence instance URL (e.g., `https://your-domain.atlassian.net/wiki`)
Cloud Instance	Toggle on for Cloud, off for Server/Data Center
Space Keys	Comma-separated space keys to sync (optional)
Page IDs	Comma-separated specific page IDs to sync (optional)
CQL Query	Custom CQL to filter content (optional)
Labels to Skip	Comma-separated labels to exclude (optional)
Batch Size	Pages per batch (default: 50)

GitHub — Issues, PRs, and Files

Syncs issues, pull request discussions, comments, and selected repository files from GitHub.com or GitHub Enterprise Server. Defaults to Markdown and YAML files when repository file indexing is enabled.Authentication: Personal access token, or a GitHub App configured under Settings > GitHub Apps.

Field	Description
GitHub API URL	API endpoint (e.g., `https://api.github.com`)
Owner	GitHub organization or username
Authentication Method	Personal access token or GitHub App
Repositories	Comma-separated repository names (optional — leave blank for all org repos)
Include Issues	Sync issues and their comments (default: on)
Include Pull Requests	Sync pull requests and their comments (default: on)
Include Repository Files	Sync repository files (default: off)
File Types	Comma-separated file extensions when repository files are enabled (defaults: `.md`, `.mdx`, `.yaml`, `.yml`)
Labels to Skip	Comma-separated labels to exclude (optional)

Notion — Pages and Databases

Syncs pages from a Notion workspace.Authentication: Notion integration token (starts with secret_). Create an internal integration and share the relevant pages or databases with it.

Field	Description
Database IDs	Comma-separated Notion database IDs (optional — leave blank for all accessible pages)
Page IDs	Comma-separated specific page IDs (optional — takes precedence over Database IDs)

SharePoint — Documents and Site Pages

Syncs documents and site pages from SharePoint Online. Supported types include .txt, .md, .csv, .json, .xml, .html, .yaml, .docx, .pdf, and .pptx. Image files up to 4 MB are indexed when a multimodal embedding model is configured.Authentication: Azure AD app registration with Sites.Read.All application permission on Microsoft Graph (admin consent required).

Field	Description
Tenant ID	Your Azure AD (Entra ID) tenant ID or domain
Site URL	Your SharePoint site URL
Client ID	Azure AD app registration Application (client) ID
Client Secret	Azure AD app registration client secret value
Drive IDs	Comma-separated document library IDs (optional)
Folder Path	Restrict sync to a specific folder path (optional)
Recursive	Traverse subfolders (default: on)
Include Pages	Sync site pages and web part content (default: on)

Google Drive — Files and Workspace Docs

Syncs files from My Drive and Shared Drives. Supports .txt, .md, .csv, .json, .xml, .html, .yaml, .docx, .pdf, .pptx, Google Docs, Sheets, and Slides. Image files are indexed when a multimodal embedding model is configured. Files larger than 10 MB are skipped.Authentication: Service account JSON key (recommended) or a short-lived OAuth2 access token with drive.readonly scope.

Field	Description
Drive IDs	Comma-separated shared drive IDs (optional — leave blank for My Drive)
Folder ID	Restrict sync to a specific folder (optional)
File Types	Comma-separated extensions (optional — leave blank for all)
Recursive Traversal	Sync files from all nested subfolders (default: on)

Salesforce — CRM Records

Syncs CRM records from a Salesforce org. Defaults to Account, Contact, Opportunity, and Case.Authentication: Salesforce username, password, and security token. Enter the password directly concatenated with the security token in the Password + Security Token field (no separator).

Field	Description
Login URL	Salesforce login endpoint (default: `https://login.salesforce.com`)
Email	Your Salesforce username
Password + Security Token	Password concatenated with security token (e.g., `MyPassword123XXYYZZ`)
Objects	Comma-separated Salesforce object API names (optional — leave blank for defaults)
Advanced Object Config JSON	Optional JSON for precise field and association control

Example advanced config:

{
  "Lead": {
    "fields": ["FirstName", "LastName", "Company", "Email"],
    "associations": { "Account": ["Name"] }
  },
  "Case": {
    "fields": ["Subject", "Status", "Priority", "Description"]
  }
}

Other Supported Connectors

Archestra also ships connectors for:

GitLab — issues and merge request discussions from GitLab.com or self-hosted instances
Asana — tasks and user comments from selected Asana projects
ServiceNow — incidents, change requests, problems, and business applications
OneDrive — files from OneDrive for Business personal drives (work/school accounts only)
Dropbox — text and source files from Dropbox accounts or team folders
Linear — issues, projects, and cycles from a Linear workspace
Outline — published documents from an Outline workspace

Each connector is configured from Knowledge > Connectors > New Connector.

Assigning a Knowledge Base to an Agent

Open the agent

Go to Agents in the left sidebar and click the agent you want to attach knowledge to, or create a new one.

Open Knowledge Sources

In the Edit Agent dialog, scroll to the Knowledge Sources section.

Select sources

Click Select connectors or knowledge bases and pick one or more entries from the Knowledge Bases and Connectors lists. An agent can be assigned multiple Knowledge Bases and individual connectors simultaneously.

Save

Click Update to save. The agent now has a query_knowledge_sources tool that searches across everything attached to it.

Managing Connectors

Open any connector from the Connectors page to:

Toggle enabled/disabled — suspends or resumes the cron sync schedule
Trigger sync — runs an immediate sync outside the schedule
View indexed documents — browse, preview, and delete documents produced by the connector
View runs — see sync history with status, document counts, and any errors

Get Started

MCP

Agents

LLM Proxy

Security

Administration

Integrations

Contributing

Agent Knowledge Bases: Built-In RAG Retrieval Pipeline

Configuration

Embedding Configuration

Reranking Configuration

How Retrieval Works

Creating a Knowledge Base

Files (Static Uploads)

Connector Visibility

Supported Connectors

Assigning a Knowledge Base to an Agent

Managing Connectors

​Configuration

Embedding Configuration

Reranking Configuration

​How Retrieval Works

​Creating a Knowledge Base

​Files (Static Uploads)

​Connector Visibility

​Supported Connectors

​Assigning a Knowledge Base to an Agent

​Managing Connectors

Configuration

How Retrieval Works

Creating a Knowledge Base

Files (Static Uploads)

Connector Visibility

Supported Connectors

Assigning a Knowledge Base to an Agent

Managing Connectors