NOAN - The fact layer for your AI agents
by•
Your company has thousands of documents, but only one version of the truth. NOAN turns the facts your business has approved such as pricing, positioning, policies, products, customers, and more into a verified, versioned source of truth available through API and MCP. Connect any model, agent, or app and give them all the same company facts, instead of letting each AI interpret your documents differently.
Replies
Hey Product Hunt 👋 I'm Neal, founder of NOAN.
Why we built this Before NOAN I spent years helping companies connect strategy, brand and execution. The hard part was never the ideas. It was getting everyone to work from the same version of the truth.
AI made that problem worse, fast. Every team started pointing its own tools and agents at the same scattered docs, decks and prompts: pricing in one place, positioning in another, the ICP in a prompt someone wrote months ago. Each one came back with a slightly different company. More context doesn't fix that. More retrieval just gives agents more things that look right. What they need is a list of things that are true, verified by someone who knows. So we built NOAN around that idea from day one, and we run our own company on it.
How NOAN works NOAN is a fact layer: your company's verified truths, served to people and agents from one place. Facts live in blocks (pricing, ICP, brand tone, refund policy). A person verifies each one or gives an agent the rights to update them, and every change is versioned.
Then there's Activity: a social feed of everything happening across your API. Every read, write and task, by every person and agent, in one stream. You can watch your agents work like you'd scroll a timeline, and trace every answer back to the fact it came from.
There are three ways to use it:
🔹 NOAN App. Create and verify facts, keep your contacts, and supervise agent work on a task board. No code. 🔹 NOAN Headless. Your facts as an API: versioned, permissioned, addressable by block. Build your own agents, apps and even your website on top, and connect Claude, Claude Code, Cursor or Codex to our hosted MCP server in about a minute, no API key needed. Our own site has no CMS and no CRM. It reads NOAN.
🔹 Verity in Slack. Verity is our agent, and she lives where your team already talks. Ask her anything in any channel and she answers from your verified facts. Type "fact:" and she drafts one for the right person to verify. If someone says something that contradicts a fact, she flags it. She also adds tasks, contacts and notes, turns meeting notes into proposed facts, and can manage all your own agents from the task board.
How it's different Most teams building a company brain today are doing it in markdown files or a GitHub repo. It's a good start, but a folder of files can't tell you which version is true, who verified it or what each agent read. A company needs a connected graph of its facts, contacts and work, behind an API every person and agent can reach.
Most tools come at this from the wrong end: ❌ Workspaces and wikis are written for people, so agents retrieve whatever looks similar ❌ Memory and RAG recall what was said, not what the company has agreed is true ❌ Agent platforms sell you the agents, meter every task and keep the brain locked inside
NOAN makes building accurate agents simple. So we're giving ours away.
New today: the agent pack, MIT licensed The agents we actually run NOAN on, including customer support, fact alignment, market research, weekly activity reports, sales decks and our newsletter, with more joining as we build them. Bring your own model keys, run them on your own GitHub, no per-task meter. Read the code, fork it, make it yours.
🎁 For the Product Hunt community 50% off NOAN Starter for your first 3 months. Use code PRODUCTHUNT50 at checkout.
🚀 Get started
Sign up free at getnoan.com. Verity reads your website and drafts your first facts for you to verify.
Connect your Agent: api.getnoan.com/docs/
Add Verity to your Slack: slack.getnoan.com/install
Clone the agents: github.com/getnoan/agent-pack
This is a relaunch for us. We first launched here in 2025 as a tool for entrepreneurs, and our users pulled us somewhere more interesting. I'll be here all day and would love your hardest questions, especially this one: where should a fact layer stop and memory start?
@neal_mann nice launch! the variance between company memory and verify truth makes the lot of sense.
how much the setup does it usually takes before an agent can start using the NOAN competently.
@amrita_prajapati2 thanks! It's very quick, you just give your AI platform of choice, say Claude, the API docs and it sets you up on NOAN. It will help you build your fact layer and it then understands how to use all the functionality, like the task board or contacts network as it's all on the API - and it gets going!
@neal_mann @amrita_prajapati2 Our onboarding will set you up with your first facts - taken from your website, then you can connect to any LLM to suggest facts based on what it remembers about your business and by asking it to export facts from documents.
Not gonna lie didn’t see the point of getting this when I already had Claude but now with the API and MCP it’s a game changer. To have a single source of truth is so refreshing both for me and especially for my team who aren’t as up to date on Ai
@luc_kelly exactly, for instance you can have a non-technical team using claude cowork etc and on the same fact layer as the technical team building your customer support agent. It just takes a mindset shift of what your company is - it's about fact control on the API!
@luc_kelly Thanks for the support!
Hey Product Hunt 👋 Dan here, one of the makers.
Once we released the NOAN API and MCP, something changed for us: we could build any agent to help run the business in minutes. We now have 27 agents performing different automations.
Every agent we run is the same small program running on a schedule in GitHub Actions.
Each agent is guided by facts in NOAN. One fact is its playbook for what to do and how to decide. Another sets its tone and templates. Both sit next to the facts they work from: our pricing, positioning and customers.
So when we want an agent to behave differently, we edit a fact, not code. The next run picks it up.
That's how we run our sales, customer success, operations and weekly reporting today.
Happy to answer questions!
The business brain finally has a home instead of living in 19 Google Docs and a group chat haha. Cool idea, congrats on the launch! @neal_mann @danmonthell @imfilp
@danmonthell @imfilp @zach_francis haha It does indeed! Thanks a lot.
Congrats on the relaunch Neal and team! Full disclosure, we work with Noan, but the Activity feed is the bit I'd point people to - being able to trace an agent's answer back to the exact fact it read is what makes it safe to let agents near sales and client comms.
Great to see the agent pack open sourced too.
the fact-level permissions and drift detection answer above is solid for async channels where someone can wait for approval. our case is voice, the agent is mid-call and has maybe 800ms to answer, so there's no time to check for a pending approval. curious whether NOAN distinguishes "verified and stable" facts from "approved but still drafty" ones at read time, so a live agent could be configured to only pull the former and fall back to "let me check on that" for the rest
@galdayan we use voice too! Our voice agent runs on the fact layer, our investors talk straight to her for instance too. Facts are either live or draft, it's about what is live on the API. So a voice agent would be a service using the API, as a result it's what's live that matters to it. It's a different way to think about how a business is run, your facts control the company and agents. If given write permissions it can also update facts or the contacts details it is talking to as well in real time.
@galdayan facts are approved decisions. Our agents only ever answer questions if it can find the answer in our fact base. It will say outright that it doesnt have the answer if it cant find it. At which point a human gets flagged in the task manager so we can fill in the gap for next time. Our agents will never provide an answer that is not recorded as a fact.