Projects Overview
Overview
Projects in Inkeep let you group together related sources and settings specific to a scenario.
For example, you may have independent projects for:
- Customer-facing (Public) vs employee-facing (Private) assistants
- Independent products within your organization
- Development vs production environments
When you start with Inkeep, you'll typically already have at least one project pre-made for you with all your public sources already selected.
Public vs Private Projects
We recommend separating your projects based on their audience to ensure appropriate content access. Typically, this means two projects for your team vs your users:
| Type | Use Case | Sources | Integration types |
|---|---|---|---|
| Public | Customer-facing assistants | Public | - Website chat - Documentation - Community forums |
| Private | Employee-facing copilots | Public & Private | - Support Copilot - Internal Slack channels - Help desk systems |
Settings
Projects have several key settings that define their scenario. These can be configured under Settings under a project in the Inkeep Dashboard.
Chat Subject Name
The topic that the AI Assistants in the project are knowledgeable about. Typically, this corresponds to a product name like Inkeep.
Get Help Links
When the AI assistant can't answer, it'll incorporate in it's answer links to your support channel[s].
For example, the AI assistant may say I wasn't able to find information about {X}, please reach out for additional help at {get_help_link}
Support channels can include:
- Contact us pages
- Help Center
- Community forums (Slack, Discord)
- Email address
Custom Guidance
These are custom instructions (prompts) that extend the default behavior of the AI assistant to modify its tone or behavior.
Check out this tutorial to learn more.
Manage Sources
A Source is a collection of content that your AI assistant can reference. A project can use multiple sources, and sources can be shared between projects.
- Go to the Inkeep dashboard
- Select Projects and click on the project you'd like to configure
- Click on Sources
- Click on Select Sources
- Check or Un-check sources to modify the ones used by the Project.
- Click Save
To create additional projects or modify sources, please contact us.
Integrations
Integrations are basically AI assistants for a given project. Integrations generally use and inherit settings from the project.
Types
- Web: For your website, docs, app, or help center. See our quickstarts.
- Slack: A bot for Slack communities or workspaces.
- Discord: A bot for Discord communities.
- Support Copilot: AI assistant that works as a sidebar within help desks.
- API: For custom UIs or workflows that use the Inkeep AI API (chat completion APIs).
Creating an Integration
- Click Create Integration in your project
- Select the integration type
- Configure integration-specific settings
- Get your API key and follow the relevant guide
Each integration's usage can be tracked separately in analytics.
FAQs
While it is up to your company's own policies and best practices, chat and search API keys are generally ok to be included in client-side (browser) code if the search and chat service is intended for public, unauthenticated use. This is typical of public-facing documentation sites or marketing pages. We implement various protection mechanisms to prevent against abusive usage and in general the chat service is scoped to answering questions about your product so is not a general use API key like an OpenAI or general LLM provider API key would be. If you would like to implement your own protection mechanisms, it is possible to have traffic routed through a private proxy, implement CAPTCHA, and/or add user authentication.
Teams typically create an Integration within a project they use for staging environments. This can help separate out usage from your production usage. If you need truly independent staging or testing environments, you can set up two projects that mirror each other instead.
Separately, if using our UI components, you can set baseSettings.env to DEVELOPMENT so that questions asked in that mode are not shown in the analytics dashboard.