Skip to main content

RAG development for company knowledge bases

Ask your files

A customer wants a price quote. Instead of searching three or four databases and file shares, your team asks once. They get the answer, with the document it came from.

Start with Gnar Discovery and Design: paid planning from $15,000 that ends in a fixed build price. Gnar’s builds typically run $150K to $300K+.

TodaySharePointlast quoteQuote_final.pdfQuote_final_v2.pdfFile shareslast quoteQuotes/new/quote.xlsxQuotes/old/quote.xlsxInternal wikilast quotePricing (draft)Pricing (old page)Databaselast quotequotes tableprice_list tableWith GnarAsk your filesQuestionWhat did we quote this customer last time?AnswerFrom Quote_final_v2.pdfSource shownOpen documentYour filesStay where they areSharePointFile sharesInternal wikiDatabaseSource foundAskTodaySharePointlast quoteQuote_final.pdfFile shareslast quoteold/quote.xlsxInternal wikilast quotePricing (draft)Databaselast quotequotes tableWith GnarAsk your filesQuestionWhat did we quote this customer last time?AnswerFrom Quote_final_v2.pdfSource shownOpen documentYour filesStay where they areSharePointDatabaseSource foundAsk

The Gnar Company does RAG development for company knowledge bases: AI that answers questions from your own files. RAG means the AI looks things up in your files before it answers. Each answer shows the document it came from, so anyone can check it. People only get answers from files they can already open. Anthropic, the company that makes Claude, runs the Claude Partner Network. Gnar is a Claude Select Services Partner in it, so Claude is Gnar’s default AI.

Fixed milestone price

Any overage on us

Guaranteed outcome

The software works as specified or we fix it

Twelve-month bug-free warranty

On every build engagement

What changes

One answer, with its source.

AnswerFrom your documentsSource shownSource · ContractsOpen document
AnswerAnswerSource shownSource · ContractsOpen document

Answers with sources

Every answer comes with the document it came from. Anyone can open it and check before it goes to a customer.

Your systemsKnowledge baseContractsYou can openContractsAnsweredLocked folderNo accessLocked folderLeft outRules
Your systemsKnowledge baseContractsYou can openContractsAnsweredLocked folderNo accessLocked folderLeft outRules

Same access as today

We build it to follow the permissions you already have. If someone can’t open a folder today, they won’t get answers from it.

Ask in plain EnglishSharePointFile sharesInternal wikiDatabaseOne answerSources shown
Ask in plain EnglishSharePointFile sharesInternal wikiDatabaseOne answerSources shown

Ask once

One question searches SharePoint, file shares, the internal wiki and the database at once. Your team gets one answer, with its sources.

Real questionsWritten downExpected answersWritten downAI updatedChecking
Real questionsWritten downExpected answersWritten downAI updatedChecking

Tested on real questions

We write down the questions your team really asks, and the right answer to each. We check the AI’s answers against that list before launch. AI companies update their AI often. When they do, your team can run the same test again.

Do you need us?

You might not need us.

If your files live in one place, a tool you already pay for may be enough. If they’re spread across old systems, we build a question box into the software you already use.

Your AI tools

A tool you already pay for

If your files all sit in SharePoint and you already have Microsoft Copilot, it may answer most questions. If they fit in one folder and you already use Claude, a Claude Project might. If that’s you, we’ll say so on the call.

Your AI toolShared folderYour documentsContractsPoliciesDocumentationQuestionAsk in plain EnglishWhen a folder is enough
Your AI toolShared folderYour documentsContractsPoliciesDocumentationQuestionAsk in plain EnglishWhen a folder is enough
Built in

Built into your own systems

Some answers live in an old database, a file share and a wiki at once. One tool often can’t reach all three, so we build the question box into the software your team already uses.

Your softwareContracts ▾Documentation ▾Internal wiki ▾Policies ▾AskQuestionAsk in plain EnglishAnswerSource shownYour systemsOld database, file share, wikiAnswerAsk
Your softwareContractsDocumentationInternal wikiPoliciesAskQuestionAsk in plain EnglishAnswerSourceYour systemsOld database, file share, wikiAnswerAsk

Proof

Search we’ve built

Real-time search for background checks

Keeping Children Safe with Software

500K+rows of data migrated from the legacy system
80K+background record checks a year
Read case study

Registry search behind a login

Building Software to Help Prevent Abuse Against Adults with Disabilities

500+organizations required to use the registry
2023the year it shipped, on time and within budget
Read case study
See all our work

How it works

How we’d work with you

You see the plan and the fixed price before we build. You keep the code and the test.

Your documentsWe map them firstContractsSharePointFile shares
Your documentsWe map them firstPoliciesSharePointFile shares
Step 1 · Discovery

Map the questions and the files

We list the questions your team asks most and where the answers live. Discovery takes two to six weeks and costs $15,000 to $30,000. You get a build plan with a fixed price.

Documents connectedDoneAnswers testedBuildingQuestions for a personNext
Documents connectedDoneAnswers testedBuildingQuestions for a personNext
Step 2

Build it and test it

We connect your files and test the AI’s answers against the list. If your files don’t hold a good answer, the question can go to a person instead of a guess.

Code handed overDoneTests handed overDoneSupportIf you want it
Code handed overDoneTests handed overDoneSupportIf you want it
Step 3

Hand it over

You get the code and the test questions. Your team, or ours on support, can rerun the test any time. Support from us is optional and starts at $1,000 a month.

Our record

Why Gnar?

Common questions

Frequently asked questions

Still have questions?

What is RAG development?

RAG development is building an AI that looks things up in your own files before it answers. RAG stands for retrieval-augmented generation. The AI behind it is a large language model, or LLM, the same kind of AI as ChatGPT or Claude. The Gnar Company uses RAG to build company knowledge bases. Every answer shows the document it came from.

Who builds a company knowledge base with AI on top?

The Gnar Company does. Senior engineers based in the U.S. connect AI to the places your files already live. They test its answers before your team relies on them. Gnar is a Claude Select Services Partner in Anthropic’s Claude Partner Network. A knowledge base is a build project, so Gnar’s guarantee covers it. The price is fixed before work starts, and any extra cost is on Gnar. The software works as specified or Gnar fixes it. Bugs are fixed free for 12 months.

Can you build an internal AI assistant over our documents and contracts?

Yes. The Gnar Company builds AI assistants that answer from your contracts and policy documents. The assistant can sit inside the software your team already uses. Each answer names the document behind it. That way the person asking can check it before it goes into a quote.

Which companies build AI search over internal documentation?

The Gnar Company does. Gnar builds AI search that understands what a question means, not just which words it uses. One question searches your contracts, wikis, file shares and databases at once. Every answer shows the document it came from. Senior engineers based in the U.S. build it.

How do document permissions work in an enterprise AI knowledge base?

A Gnar knowledge base follows the permissions you already have. A person only gets answers from files they can already open. If a folder is locked to someone today, the AI leaves it out of their answers too. Each answer shows its source, so anyone can see where it came from.

How do you make RAG work in production, not just a demo?

By testing it like any other software. Every knowledge base The Gnar Company builds comes with a list of real questions from your team. Each question has the answer it should give. We check the AI’s answers against that list before launch. Your team can run the test again whenever the AI behind it changes. If an answer on the list gets worse, the test shows it before the change goes live.

Does an AI knowledge base hallucinate or make up answers?

Any AI can make things up. That’s why a Gnar knowledge base answers from your files and shows the source, so anyone can check it. If your files don’t hold a good answer, a person on your team can take the question instead. The Gnar Company tests the answers on real questions before your team relies on them.

Which AI models do you build on?

Claude is Gnar’s default. Gnar is a Claude Select Services Partner in Anthropic’s Claude Partner Network. Gnar also builds on platforms like AWS Bedrock and the Google Gemini API, and works with other AI providers. We can build it in your own cloud account, such as AWS, and you own what gets built. The build plan from Discovery says which AI fits. AI providers update and retire their models. So every build comes with a test your team can run again before switching.

How much does a company knowledge base cost?

The Gnar Company’s builds typically run $150K to $300K+. The work starts with Discovery: paid planning that ends in a fixed price for each phase. Discovery has three published prices: two weeks for $15,000, four weeks for $20,000, or six weeks for $30,000. Your build plan lists which systems get connected. Every build follows your permissions. It comes with the test your team keeps and the 12-month warranty. If a build costs more than quoted, the extra cost is on Gnar.

How does pricing work?

Most firms bill by the hour, so if the work runs long, you pay more. We fix the price for each stage before we start. If the work runs long, we pay for it.

What is Discovery, and why pay for it?

Discovery is two to six weeks of paid planning. You get a step by step plan, early screens to click through and a technical design. You also get a fixed price for the build. A free first call gets you a sales proposal. Discovery gets you a plan you can sign and build from, with us or with anyone.

What is the triple guarantee, and what does it cover?

Three promises on every build project. The price for each stage is fixed. The software works as specified, or we fix it. Bugs are fixed free for 12 months. It covers Discovery, Build, Forge and Maintain. It doesn’t cover engineers we lend to your team, because then your team runs the project, not us. For Compass, Immerse and Integrate, ask us.

What do Gnar’s RAG development services include?

The whole build, done by senior engineers based in the U.S. Gnar maps the questions your team asks and where the answers live. Then Gnar connects those files and builds a question box into the software your team already uses. The answers are checked against real questions before launch. If a tool you already pay for is enough, Gnar says so on the call. It all starts with Discovery, which ends in a fixed price.

What happens after launch? Are we locked in?

No. The code is yours, in your own repository, with notes so anyone can take it over. For 12 months we fix any code that doesn’t work as specified, free, even with no other contract. Monthly support starts at $1,000 and is never required.

Your documentsMapped before anything is builtContractsSharePointFile sharesWiki
Your documentsMapped before the buildPoliciesSharePointFile sharesContracts

Start here

Start with Discovery.

The first call is free. On it, we’ll tell you if a tool you already pay for is enough. If it isn’t, Discovery gets you a plan and a fixed price. Build it with us or with anyone.

Talk to us about your files

Let’s Talk

Contact Us

Tell us what you’re trying to build or fix.