What is Baseera
Baseera is an Arabic-first platform for AI assistants that answer from your documents, act in your app only with approval, and are tested before release.
Baseera (بصيرة) lets you put an AI assistant inside your own product. You configure it in a dashboard, install an SDK in your app, and the assistant appears as a chat panel your users can talk to.
Three rules shape everything else:
- It answers from your documents. Replies cite the sources they used, and users can open the original when you allow it.
- It acts only with approval. Tools run in your app, with the signed-in user's own session. A tool that changes data waits until the user presses Approve.
- It is tested before release. You save real conversations as test cases, run them on a draft, and publish only when they pass.
Who it is for
Baseera is for teams that build a web app and want an assistant in it, for example an invoicing tool, a customer portal or an internal back office. You write your tools in your own codebase, and you bring your own model key (Anthropic, OpenAI, Google or OpenRouter). Arabic and right-to-left layouts are first-class: the dashboard, the chat panel and the approval cards all ship in Arabic and English.
The shape of the product
Knowledge
Upload files, crawl a website or push documents from your server. Baseera splits, embeds and searches them for you.
Agents
A system prompt, a model, a tone and the knowledge to use. Edits go to a draft until you publish.
Tools and approvals
You define tools in your app. Reads run at once. Writes and drafts wait for the user.
Evaluation
Test cases with mocked tools and typed checks, an optional model judge, and a gate on publishing.
Embed
Add a provider and a chat surface to a React app, and sign a user token on your server.
Immersive mode
Chat is one way to work with an assistant. The other is immersive mode: the assistant sits beside the record or form a person is working on in your app, for example an invoice. Your app declares the record's fields and its own checks, and Baseera adds AI checks that read your knowledge. A problem is flagged on the field that is wrong, the model drafts a fix the person approves in the field, and blocking issues hold the Finish button. Every flag, fix, approval, override and finish is kept as a trail. Both modes can be on for one agent. See Immersive mode.
What you do, in order
- Add your model provider key in the dashboard.
- Create an agent and attach knowledge.
- Install the SDK, register your tools and add the chat surface.
- Test on the draft, then publish.
- Watch sessions in production, with every message, tool call and approval in a trace.
The SDK packages (@baseera/core, @baseera/react, @baseera/node, @baseera/ui) are MIT licensed, but they are not on npm yet. The quickstart explains how to run them from the repository until the first release.