Haystack expands AI write access to ChatGPT, Copilot, and Slack

From search to creation

Haystack, a platform designed to centralize internal company communications, is expanding how employees interact with their company's private data. While the platform previously allowed users to find information, it is moving into a phase where users can actively create new content through AI interfaces.

The company is expanding its Model Context Protocol (MCP) server. MCP is an open standard that allows different AI models to connect to external data sources and tools. By using this standard, Haystack ensures that companies are not locked into a single AI provider, allowing them to use the tools their teams already prefer.

The service is now available in OpenAI's ChatGPT, Microsoft Copilot Studio, and the Slack Marketplace, joining Anthropic's Claude, which received the update in August. Starting October 26, these platforms will support 'write access,' a feature that lets the AI perform actions like drafting and saving content rather than just answering questions.

How the workflow functions

The integration works by connecting Haystack to an organization's internal systems. Once an administrator connects the platform, employees sign in with their own unique accounts. This ensures that the AI assistant respects the specific permissions of each individual user. If an employee does not have permission to see a certain folder or database, the AI will not show that information to them.

When an employee uses the write feature, the AI assistant can draft a knowledge page and save it as a private draft. It can also generate a post for a specific group that the employee has the authority to publish to. Because the assistant can also connect to external systems like Salesforce, it can pull in real-time data to create summaries.

For example, a sales leader could ask the AI to turn the previous quarter's numbers into a company-wide digest. The AI would gather the data from the sales system, format it into a post, and prepare it for publication. This process is designed to happen in minutes, significantly reducing the manual effort required to keep a large company informed.

Security and permissions

A primary concern for large companies using AI is data privacy and the risk of information leaking to the public. Haystack addresses this by ensuring that every connection is authenticated as an individual employee. The AI assistant is restricted to the 'sight' of the person asking the question; it cannot see data that the human user is blocked from viewing.

The company maintains several security certifications, including SOC 2 Type 1 & 2, ISO 27001, and HIPAA compliance. These are standard frameworks that verify how a company handles data security and privacy. Furthermore, Haystack explicitly states that no customer data is used to train the underlying AI models, which is a critical requirement for many enterprise environments.

Administrators also maintain a level of control over the deployment. They can choose to block specific tools within their AI assistants if they do not want certain features enabled. This allows for a granular approach to how much power the AI has over the company's internal communications hub.

The roadmap for companies

The rollout of these features follows a deliberate expansion strategy. By starting with Claude in August and moving to ChatGPT, Copilot, and Slack, Haystack is aiming to meet employees where they already spend their time. Instead of forcing workers to log into a separate website to find information, the information is brought to their existing chat and productivity apps.

The goal is to create a 'single source of truth' for company knowledge. By allowing both the finding and the publishing of content to happen in the same place, the platform aims to reduce the friction of internal communications. This is particularly useful for large organizations that struggle to keep information current across different departments.

Companies like Weight Watchers, Thumbtack, and NerdWallet currently use the platform to manage their internal news and knowledge. The new write capabilities are expected to further streamline how these brands distribute information to their thousands of employees, making large organizations feel smaller by making information more accessible and easier to produce.

Next steps for users

The connectors for ChatGPT, Copilot, and Slack are available to Haystack customers immediately. However, the ability to write and publish content will not be active until the October 26 launch date. Organizations looking to test the system can schedule a demo to see the permissions and write features in a live environment.

For those already using the platform, the transition will involve identifying which data sources, such as Salesforce or internal wikis, should be connected to the MCP server. This will determine what the AI can 'read' and eventually 'write' to during the October update.