AgentConnect launches open-source multi-agent collaboration platform

AgentConnect launched an open-source, self-hostable platform on August 25, 2026, to let teams bring multiple AI agents into the conversations and tools they already use. The product aims to move agent work out of isolated terminals and into shared workflows across Slack, GitHub, Telegram and Discord.

Why it matters: - AgentConnect targets a growing gap in AI agent use: agents can already do more work, but team collaboration often still happens in private, fragmented sessions. - The platform is built to keep agent execution and workspaces inside a team-operated environment, which may matter for organizations that want more control over access and data. - Teams can use multiple ACP-compatible agents together instead of forcing every workflow through one bot or one provider.

What happened: - AgentConnect introduced an open-source platform for teams and multiple AI agents to work together across existing communication and development tools. - The launch took place on August 25, 2026, in San Francisco. - The platform is designed to bring agents into existing team conversations rather than move work into a separate workspace. - AgentConnect supports Claude Code, Codex, and other ACP-compatible agents. - Team members can assign distinct roles to different agents and let people and agents collaborate through shared conversations and agent-to-agent calls.

The details: - In incident response, a team can bring in a triage agent to review context, a code specialist to trace a regression, and a verification agent to check the result. - Human teammates can add context, redirect work, and follow decisions in the same Slack thread. - A support issue can move from a customer-facing Telegram conversation into a trusted Slack workspace without manual copying. - Engineering teams can use general-purpose and specialist agents to review pull requests based on risk level. - Agents can leave inline GitHub comments and produce independent check results. - Scheduled work and webhook-triggered tasks can continue in a shared conversation when human input is needed. - Each agent can have separate models, workspaces, memory, MCP servers, skills, repository access, and sandbox policies. - AgentConnect connects to Slack, Telegram, and Discord. - Work can start from conversations, GitHub pull requests, webhooks, or schedules. - Teams can manage agent access separately from session visibility and control which tools, repositories, and other agents each agent can access. - Multiple social accounts can be linked to one member profile so supported platform permissions apply consistently. - AgentConnect is licensed under Apache 2.0 and can be deployed with Docker Compose or Kubernetes. - Source code, product information, and documentation are available through the AgentConnect website.

Between the lines: - The launch reflects a push to make AI agents behave more like teammates than isolated tools. - Shared conversations and agent-to-agent calls could reduce context loss and make handoffs easier across support, engineering, and incident response workflows. - The self-hosted model positions AgentConnect for organizations that need tighter operational control than many hosted agent products provide.

What's next: - Teams can now adopt AgentConnect OSS and test it in their own infrastructure. - The platform is likely to compete on workflow flexibility, access control, and self-hosting rather than on a single-model experience. - AgentConnect is directing users to its website for code, product details, and documentation.

The bottom line: - AgentConnect is betting that the next step for AI agents is not more autonomy in isolation, but better collaboration inside the tools teams already use.