Company documents
Add your employee handbook, safety policies, operational procedures, and company guidelines as entries. Agents can instantly answer questions like “What is our refund policy?” or “How do I submit a leave request?”
Knowledge bases are how your agents know about your business. They are searchable collections of information — company policies, product details, process guides, customer data, and anything else your agents need to answer questions accurately and make good decisions.
A knowledge base is a searchable index of entries. Each entry has a title, content, and keywords. When an agent needs information during a conversation, it searches the relevant knowledge base and uses the results to inform its response.
Without knowledge bases, agents can only rely on their general training and the persona you wrote. With knowledge bases, they have access to your organisation’s specific information — the details that make their answers accurate and useful.
Every Sprigr Team organisation organises its knowledge into a set of standard indexes. Shared Knowledge, Admin Data and the Agent Directory are set up when your organisation is created. If you described your business at signup, Shared Knowledge starts with a Company Context entry built from that description, and the Agent Directory fills in as agents are created.
The main company-wide knowledge base. This is where you put information that any agent might need:
Every agent in the company can search Shared Knowledge during conversations, and agents save company-wide facts here using the sprigr_write tool.
A platform-managed index (badge Hive on the Knowledge page). It appears in the Knowledge grid alongside the other standard indexes, and owners and admins can review, edit, and delete its entries like any other (other members can change only entries they wrote) — see Managing Indexes. It is not the path by which agents share what they learn; that is the consolidation pass described below.
Facts an agent picks up during a conversation, or from a task another agent delegated to it, are captured as pending memories. They stay private to that agent unless a verified member of your team explicitly asked for the fact to be shared with everyone.
Each memory carries attribution fields that sprigr_write accepts alongside the title and content:
kind — the epistemic type: fact, take, bet, hunch, preference, commitment, event, or beliefholder — who asserts the claim (the user, a named person or organisation, or the system for the agent’s own analysis)subjects — what the entry is about, as typed slugs, primary subject firstweight — calibrated confidence from 0 to 1, which drives how results are rankedAbout once an hour, a consolidation pass reviews the pending memories. Short-lived ones are dropped. Durable facts, preferences and beliefs are kept: most go into the capturing agent’s private notes, and only those a verified team member asked to share go into Shared Knowledge. Duplicates are merged, changed facts replace the old version, and a per-subject summary is kept up to date. A memory still pending after 30 days expires. Shared entries show up on the Knowledge page like any other Shared Knowledge entry.
An owner- and admin-only index (badge Admin on the Knowledge page) for material that should not be visible to every member. Only owners and admins can browse it.
An automatically maintained index of all agents in your organisation — who they are and what they can do. Other agents (and your companion agent) use this to figure out which specialist to delegate a task to.
You do not need to maintain this manually. The platform updates it as you create, modify, and remove agents.
Each agent also has a private index for its own notes and reminders (agents choose the private scope when saving). Other agents cannot read it. On the Knowledge page it is visible only to the people the agent belongs to: a companion’s notes to its person, and a shared agent’s notes to the members assigned to it. Admins and owners get no extra access.
When an agent searches a knowledge base, Sprigr Team uses hybrid search — a combination of two approaches:
The two approaches are combined to give agents the best of both worlds — precision when exact terms match, and flexibility when the user phrases things differently.
The Least used (30d) sort on the Knowledge page shows entries agents rarely read, so you can review or remove them.
Indexes synced from a document source that has its own permissions (for example OneDrive or SharePoint through the Microsoft 365 integration) are ACL-enforced. Every synced document records the users and directory groups allowed to see it, and every search an agent runs on behalf of a signed-in person is restricted to that person’s own identity and group memberships. Group membership comes from your SSO provider (Microsoft Entra group memberships are refreshed on each login), and a document with no resolvable owner is visible to nobody, so access fails closed rather than open.
Company documents
Add your employee handbook, safety policies, operational procedures, and company guidelines as entries. Agents can instantly answer questions like “What is our refund policy?” or “How do I submit a leave request?”
Product and service info
Add your product catalogue, service descriptions, pricing sheets, and technical specifications. Agents can help customers find the right product, provide accurate pricing, and explain features.
Process guides
Document your internal workflows — how to process a return, onboard a new client, schedule a technician, or close out a job. Agents follow these guides step by step when helping users.
Customer data
Store customer profiles, account details, and interaction history. Agents can look up a customer’s past orders, preferences, and outstanding issues to provide personalised service.
sprigr_search and sprigr_write to interact with knowledge bases.