Kontxt turns what your team reads and highlights into permissioned AI memory. Free Starter plan; Business is $20 per person a month. Features and pricing.
Researched and fact-checked by Bartu Cavusoglu · Updated October 26, 2026
Kontxt is an enterprise context engine that turns what a team reads, highlights, and discusses into memory for AI assistants and agents. The company's pitch is that AI knows a lot in general but not what matters to a specific team, and Kontxt tries to fill that gap with a record of team judgment.
Instead of storing documents alone, it records who highlighted a passage, where it came from, who it is shared with, how it is tagged, and what discussion followed. Any AI tool with permission to see that record can retrieve it and cite it in an answer.
Kontxt describes four systems. Acquire captures highlights, notes, tags, and the passive trail of what a team reads. Govern attaches identity, permissions, and provenance to every record. Select ranks what a user is allowed to see and keeps only what the task needs. Deliver hands each agent its slice through MCP and an API, with citations attached.
Capture runs in the browser today, with no integration needed, across web pages, PDFs, docs and wikis, dashboards, and discussions. Native connectors for Slack, Drive, and Confluence are listed as roadmap. Deployment is described as one line of JavaScript, and permissions are checked on each request.
Retrieval ranks by attention signals such as dwell time, revisits, and teammate engagement. Chained questions are split into steps, so a query about the cause behind a bad support month resolves in two permission-scoped lookups. Each returned block includes a stated reason for why it was supplied, which makes results easier to audit.
Users can comment, tag, run polls, and share one to one with a link to the exact spot. Business adds auto highlights, tags and summaries, plus folders, groups, and permissions. The site also describes audit views of what is read, shared, and used per page and per team, and says no customer data trains other models.
Kontxt is an enterprise context engine that turns what a team reads, highlights and discusses into permission-aware memory. It then serves that memory to AI assistants and agents such as ChatGPT, Claude, Copilot and Cursor.
Starter is free for one person with 100 AI calls a month. Business costs $20 per person a month, or $200 per person billed yearly. Enterprise is custom priced with a 25-person minimum and an annual contract.
Yes to both. The Starter plan is free, and every new account begins with 7 days of Business without a card. After the trial, the account stays on Starter unless the user upgrades.
An AI call is one piece of AI work, such as a question answered in chat, a Memory lookup from Claude, ChatGPT or another MCP tool, or an auto highlight, tag or summary. Saving, highlighting, commenting, tagging, sharing, folders and search never count.
Yes. Any MCP tool can ask Kontxt for context, and each lookup counts as one AI call. The response includes only what the requesting user is allowed to see.
Kontxt delivers context through MCP and an API to tools such as ChatGPT, Claude, Copilot, Cursor and custom agents. Capture works in the browser without integration. Native connectors for Slack, Drive and Confluence are on the roadmap.
Only AI pauses. Pages, highlights, sharing and search keep working. The team can add 1,000 calls for $40 or wait until the next month, and nothing is charged without approval.
The site states that none of a customer's data trains other models. It also says access controls are enforced inside every answer and that content is private by default.