Why do I need to set "Use Responses API" to false?

Privatemode provides an OpenAI-compatible Chat Completions API. The newer OpenAI Responses API is not part of that specification, so leave the toggle set to false. If the toggle is enabled, agent requests fail against the proxy.

Which Agents features work with Privatemode?

Agent conversations in channels, threads, and direct messages work out of the box. Features built on the default agent, such as thread and channel summarization via AI Actions, work as well; note that some of them require an appropriate Mattermost license. Tool calling through MCP depends on the selected model's tool support.

The agent does not respond. What do I check?

Verify the proxy is reachable from the Mattermost server: `curl -s http://privatemode-proxy:8080/v1/models` Run this from inside the Mattermost container if you use Docker. If the request fails, confirm both containers are on the same Docker network and check the proxy logs for attestation errors: `docker logs privatemode-proxy`

The proxy fails to start with a certificate error. What is wrong?

Your network likely uses TLS inspection. Mount a custom CA bundle into the proxy container using `-v /path/to/ca-bundle.crt:/etc/ssl/certs/custom-ca.crt:ro` and set the environment variable `SSL_CERT_FILE=/etc/ssl/certs/custom-ca.crt` in the docker run command. Also make sure you are running the latest proxy image: `docker pull ghcr.io/edgelesssys/privatemode/privatemode-proxy:latest`.

Confidential AI agents in Mattermost with Privatemode

Q: Does Privatemode retain conversation data?

No. Privatemode is designed to never learn from your data. Inference requests are processed inside a confidential computing environment and no prompts, responses, or metadata are retained after processing.

Mattermost is the open-source collaboration platform that teams self-host to keep their communication under their own control. Its Agents plugin brings AI assistants into channels and threads. With Privatemode, those agents run on cloud-grade models while every message stays end-to-end encrypted.

Introduction

Mattermost integration

Introduction

AI agents for your self-hosted collaboration platform

A Privatemode agent answering in a Mattermost thread

Self-hosted chat, cloud-grade AI

Teams choose Mattermost precisely because their conversations stay on their own infrastructure. Pointing the built-in Agents plugin at a traditional cloud LLM provider undermines that: every prompt and thread an agent reads is sent to that provider in plaintext.

Encrypted before leaving your environment

The Privatemode AI proxy runs alongside your Mattermost server and handles all traffic to the Privatemode service. It verifies the service through remote attestation, encrypts each request, and only then sends it out. By the time your data reaches the cloud, no one upstream can read it. Not even us.

Hardware-enforced on arrival

Requests are processed inside a confidential computing environment, sealed off from the host, the hypervisor, and our own operators. The full stack is open source and remotely attestable, so the guarantees are verifiable, not just promised.

Benefits

Why use Privatemode AI with Mattermost?

End-to-end encrypted AI in your chat

The local proxy encrypts every prompt before it leaves your environment and decrypts every reply as it comes back. Processing happens inside a hardware-enforced confidential computing environment on the Privatemode service. Neither the infrastructure provider nor Privatemode can access your conversations, by design.

Works with the built-in Agents plugin

Privatemode adheres to the OpenAI API specification, so it plugs into Mattermost's Agents plugin as a standard "OpenAI Compatible" AI service. No custom plugin, no code changes, just a different endpoint.

State-of-the-art models, no GPU required

Your agents get access to cloud-grade LLMs through Privatemode, all running inside a confidential computing environment. No model hosting or GPU infrastructure needed on your side.

How to get started

How to set up Privatemode in Mattermost

This tutorial assumes a running Mattermost deployment with the Agents plugin installed (it ships with Mattermost v10.3 and later). You deploy the Privatemode AI proxy next to your Mattermost server and register it as an AI service. From then on, all agent traffic is end-to-end encrypted and only decrypted inside the Privatemode service's confidential computing environment.

Get your API key

If you don't have a Privatemode API key yet, you can generate one for free here.

Start the Privatemode proxy

The proxy performs remote attestation on startup to verify the Privatemode service, then transparently encrypts all requests before they leave your server.

If Mattermost runs with Docker Compose, attach the proxy to the same network so it is reachable as http://privatemode-proxy:8080. If Mattermost runs directly on the host, drop the --network flag, publish the port with -p 127.0.0.1:8080:8080, and use http://localhost:8080 instead.

Configure the AI service

In Mattermost, go to System Console → Plugins → Agents and select Add a Service under AI Services. Configure it as shown in the screenshot:

Service name: Privatemode
Service type: OpenAI Compatible
API URL: http://privatemode-proxy:8080/v1
API Key: any placeholder value; authentication is handled by the proxy
Use Responses API: false, since Privatemode implements the Chat Completions API
Default model: the model your agents should use, for example kimi-latest

Save the configuration.

Create an agent

Open Agents from the product menu (or System Console → Plugins → Agents → Agents) and create a new agent. Give it a display name and username, such as privatemode-ai, and select the Privatemode AI service you just configured. Optionally override the service's default model and add custom instructions to shape how the agent responds.

Chat with your agent

Mention the agent with @privatemode-ai in any channel or thread and it replies in a thread. You can also message it directly, or use it through the AI Actions menu, for example to summarize a long thread.

Done!

Your team now has AI agents in Mattermost, with every prompt and response end-to-end encrypted through confidential computing.

FAQ

Frequently asked questions about using Privatemode with Mattermost

Yes. The Privatemode proxy runs next to your Mattermost server and encrypts every request before it is transmitted. On startup it also performs remote attestation to verify that the Privatemode service is genuine. Requests are decrypted only inside the confidential computing environment.