Internal search with context — trained on your data.
MergeBase transforms your organization's tables, documents, and connected data into a searchable, structured knowledge layer. This content becomes the foundation for everything your system touches: automations, dashboards, AI agents, and internal tools.
It's not just storage. It's what your system refers to when it needs to reason, respond, or route, just like a person would reference a manual, guide, or historical record before making a decision.
What Is Knowledge in MergeBase?
In MergeBase, knowledge lives in tables — native tables you create and populate directly, or tables synced in from an external source. This includes:
- Native tables you build with a custom schema (17+ field types, including Formula, Relation, Rollup, and Status)
- Uploaded files (CSV, Excel, PDF, Word, audio) parsed into structured records
- Synced data from connected systems — CRMs, databases, and SaaS apps (Salesforce, HubSpot, Google Sheets/Drive, Notion, Postgres, Snowflake, and more)
- Live Human-in-the-Loop task data, surfaced as its own queryable table
Once a table exists, its content can be searched, queried by AI agents, or referenced directly inside any workflow.
Where It's Used
Knowledge in MergeBase isn't isolated. It becomes active infrastructure used throughout your system:
AI Agents Agents reference table content when generating responses, answering questions, or summarizing content.
Workflows Workflow nodes (Start, Add Record, and others) can read from and write to MergeBase tables directly — for lookups, validation, routing, or logging.
Column Enrichment The AI Column feature lets you describe a new column in plain language (e.g. "sum totals by GL Code") and get a proposed formula with a live preview before it's added to the table.
Retrieval Methods
MergeBase supports two complementary ways of finding relevant data:
Semantic Search Finds content based on meaning, not just keywords, using AI embeddings. This helps agents retrieve accurate answers even when terms differ from what's in the table.
To improve retrieval:
- Use clear, consistent field content
- Keep related information in the same field where possible
- Mark long-form, descriptive fields (not IDs or timestamps) for AI indexing
Structured Filtering Fields marked with the Filter role can be used to narrow results by exact values — status, category, date range, or connection — alongside a semantic query. This is how you scope retrieval to "only active records" or "only this connected source," rather than a separate named search mode.
Setup Guide
Step 1: Create or Connect a Table
- New table: define your own schema from scratch
- Upload file: CSV, Excel, PDF, Word, or audio — parsed into a structured table
- Connect source: link a database or SaaS app (Google, Postgres, Salesforce, and 30+ others) with scheduled or on-demand sync
Step 2: Configure AI Search For any table, open Configure AI Search to assign each field a role — Vectorize, Filter, Display, or Ignore — so the system knows what to embed for search versus what to use for structured filtering.
Step 3: Start Using It Across the System Once configured, table data can be:
- Queried by AI agents
- Referenced in workflow nodes
- Enriched with AI-proposed derived columns
Why It Matters
Centralized logic Store rules, references, and business data in one maintainable place
System-wide usage Power AI, workflows, and dashboards with the same trusted tables
Query-ready Configure a table once, then reference it consistently across every workflow and agent
Operational clarity and consistency Everyone — and everything — works from the same source of truth