How Knowably Works
A walkthrough of what you do, what happens, and what you see at each step when using Knowably to search your company knowledge.
What does it actually feel like to use this thing?
I get that question a lot. The product pitch explains why Knowably exists. The technical articles explain how the pipeline works under the hood. This page is different. This is the walkthrough. What you do, what happens, and what you see at each step.
Get Started | What is Knowably?
Connect Your Data Sources
Your knowledge already lives somewhere. Probably in several somewheres.
You start by connecting the systems your team uses. SharePoint, Confluence, Jira, Outlook, GitHub, Google Drive, Amazon S3, websites. Plus direct file uploads for anything that does not live in a cloud service.
For Microsoft, Atlassian, and Google services, you sign in with OAuth. Click, authorize, done. No API keys to copy, no service accounts to provision. For S3 and GitHub, you enter credentials once. For websites, you give us a URL and we crawl it.
You choose exactly what to sync. Specific SharePoint folders. Individual Confluence spaces. Selected Jira projects. You are not importing your entire organization - just the parts that matter for a given use case.
After the initial connection, documents stay in sync automatically. New files get picked up. Updated files get re-processed. Deleted files get removed from the index. You set it up once and forget about it.
Watch Your Documents Get Processed
Once connected, Knowably processes every document automatically. You can follow the progress in the dashboard.
Each document goes through sync, conversion, splitting, and indexing. The interesting part from your perspective is what comes out: clean, structured text that is ready for search. A 50-page PDF with tables and footnotes becomes properly indexed passages. A Confluence page with embedded images and macros becomes searchable text.
Processing is incremental. Edit page 12 of a SharePoint document and only the affected parts get re-processed. This keeps things fast even for large collections.
You do not need to configure any of this. It just works.
Technical deep-dive: document processing
Organize into Knowledge Bases
Here is where raw data becomes useful.
Knowledge bases are collections of data sources grouped by purpose. An "Engineering Docs" knowledge base might combine GitHub repositories with Confluence spaces. "HR Policies" might include a SharePoint folder and uploaded PDFs.
One data source can belong to multiple knowledge bases without duplicating anything. Your company handbook can feed both the "HR" and the "Onboarding" knowledge base.
This is not just organizational. It is a security boundary. When a customer support agent answers questions, it should not pull from internal financial documents. Knowledge bases give you that control.
Configure an AI Agent
Agents are the AI assistants that answer questions. Each one is independently configurable.
You create an agent, give it a name, write a system prompt that defines its personality and behavior, and assign the knowledge bases it should search. That is the minimum. You can be answering questions within minutes of connecting your first data source.
If you want to go deeper, every agent has its own search settings. You control the balance between keyword matching and semantic search. You can enable query expansion so the agent finds documents even when users phrase things differently. You turn on query decomposition so complex questions get broken into focused sub-queries.
Safety settings are per agent too. PII detection, prompt injection scoring, input sanitization - all configurable based on how sensitive the use case is.
One agent for internal engineering questions. Another for customer support. A third embedded on your public website. Same platform, completely different behavior.
Agent configuration details | Developer guide
Ask a Question
You type a question. The answer streams back in seconds.
That is the user experience. Behind the scenes, a lot is happening - search, ranking, filtering, generation - but you do not see any of it. You see a chat interface. You type. You get an answer.
If you are curious about what happens between your question and the response, the technical pipeline article goes deep on that. But you do not need to know any of it to use Knowably.
Get Answers with Source Citations
Every answer includes source citations. References to the specific documents and passages the agent used to generate its response.
This was non-negotiable from day one. If an agent says "the policy requires 30 days notice," you need to verify that claim against the actual document. Citations show which data source, which document, and which passage the information came from. Click through to the original.
If the agent does not find relevant information in your documents, it tells you. It does not invent an answer to sound helpful.
Conversations are multi-turn. Follow-up questions understand what was discussed before. You can ask "what about the exception for contractors?" and the agent knows you are still talking about the same policy.
You can give thumbs up or thumbs down on any response. This feedback tracks answer quality over time and tells you which agents need tuning.
Monitor and Improve
Deploying an agent is the beginning, not the end.
I spent months on the wrong assumption - that getting answers right would be a one-time effort. In practice, documents change, user questions evolve, and what worked last month drifts. You need a feedback loop.
The dashboard gives you a real-time overview. Query volume. Token usage. Storage. Cost per agent. All in one view. You know exactly what each agent costs to operate and where to optimize spend.
User feedback shows which agents are performing well and which ones need attention. If an agent consistently gets thumbs down on a topic, you know where to focus.
From there, the cycle is straightforward. Adjust search settings. Refine the system prompt. Add a missing data source. Check the results. Repeat.
Dashboard overview | Observability details
Get Started
Connect a data source. Create a knowledge base. Configure an agent. Ask your first question.