Straight answers. Library, approvals, and price.
Straight answers on the library, approvals, and billing.
General
What is Notara?
Notara is the shared context, tools, and workflow layer for teams using AI agents. One maintained library of company knowledge, plus connected tools. Build it once, then use it from Slack, Claude Code, Cursor or any other agent. The web app is where you manage it, not where the work has to happen.
How is this different from a plain agent connection?
A typical connection often gives whoever holds the key everything the server can do. Notara uses the same rules for Slack and for outside agents. Reads run. Writes wait for a person. Deletes need an admin. Every call is logged. Access is scoped to a group, so one team's library does not show up for another team.
Do I have to use Slack?
No. Slack is one way in. You can build the library in the web app and use it only through Claude Code, Cursor or other agents.
Who is Notara for?
Teams where more than one person, or more than one agent, needs the same answers. Directors and ops leads who are tired of pasting the same company knowledge into prompts, and want agents pointed at something current.
Library & Skills
What goes in the library?
A knowledge base for a team or function. Some docs always load. Some are pulled in when the question needs them. Some never reach a model. Every collection has an index the agent reads before it answers.
Why does the library index matter?
An agent that only sees a few preloaded docs will still answer with confidence. The index lists everything that exists, including docs that were not preloaded, and how to fetch them. That is the difference between a folder of notes and a library an agent can use.
Do I have to write the library myself?
You can start from a URL. Notara crawls your site and researches the company on our keys, then scaffolds a standard eight-document structure for you to edit. Bind a document to a source and Notara re-checks it on a schedule, so the library does not quietly go stale.
What is a skill?
A saved prompt, the tools it may use, and the docs it reads first. 48 ship built in. Every published edit is versioned, so you can compare two versions and revert.
Can a skill give itself more access?
No. A skill can only tighten permissions, never loosen them. Writing a skill needs the editor role. Granting a tool privilege needs admin.
Approvals & Security
What happens when an agent tries to write?
The write is checked when it runs, not only when tools are listed. Nothing has happened yet. An admin gets an approval prompt. The agent is told to wait and not claim success. If approved, the tool runs and the result comes back. If denied, the reason comes back. Either way it is logged.
Can an outside agent do everything an admin can?
No. Outside agents are never treated like a pre-approved admin. That exception is only for scheduled tasks an admin set up. Access stacks: the tool default, workspace rules, channel overrides, then the active skill. The skill can only tighten access.
How are credentials stored?
Connector credentials are encrypted at rest with AES-256-GCM and are never written to logs or into Slack messages. For Firecrawl, Apify and Perplexity you can skip credentials entirely and run on Notara's keys.
Is my data used to train models?
No. Your library, conversations and tool output are not used to train models.
Pricing & Billing
What are fair-use AI limits?
Fair-use covers ordinary agent work on the platform. Research-heavy usage can exceed it. Higher caps are an enterprise conversation, not a metered overage bill. There is no per-call billing and no prepaid credit pack.
What happens if we need more than fair-use?
If usage is consistently above fair-use, we talk about an enterprise cap. That path also covers SSO, security reviews, custom skills and white-glove. Book a call and we will say what we can support, including the things we cannot yet.
Do outside agent reads count against fair-use?
Reading your library from Claude Code or Cursor does not run the Notara agent, so it does not count. Work you ask the Notara agent to do, in Slack or through a skill, does.
Can I cancel anytime?
Yes. No contracts. Cancel from the dashboard in one click. Your libraries and connectors are preserved for 30 days.
What does it cost monthly after the installation?
$149 a month for the platform, plus $29 per seat. Ten seats is $439/month all-in. Fair-use limits apply to research-heavy AI usage. Founding companies keep this rate for as long as they stay. Billed monthly, cancel whenever.
Is there an enterprise plan?
Yes, on a custom contract. Required for SSO, security reviews, higher caps, custom skills and white-glove. If your security team has a review process, book a call and we will walk it with them, including the requirements we do not meet yet. Notara does not hold SOC 2.
What happens if I join the waitlist instead?
You hear first when installations reopen, and the published price applies when they do. No card, and we will not promise you a date we cannot hold.
Why is it $5,000 when the site said $10,000?
Because there are no external paying customers yet. You would be among the first, and ten real installs teach us more than another month of building will. That is the entire reason, and we would rather say it than dress it up as a launch promotion. Ten installations, then it goes back to $10,000 on September 2.
Getting Started
How do I connect Claude Code or Cursor?
In the dashboard, create access for a group, issue a token, and paste the URL plus token into the client's config. The client sees only the collections that group is allowed to see.
What is the fastest way to get value?
Connect Slack, point Notara at your website so it scaffolds a first library, then fix the three or four documents it got wrong. That is usually the difference between an agent that guesses and an agent that answers.
Can I request a connector?
Yes. Email support@notara.ai with the tool. Connectors ship based on what teams ask for.
Do I need engineering help to set this up?
Not for Slack or the library. Connecting Claude Code or Cursor is a copy-paste of a URL and a token into that client's config.