Own the data boundary
Keep event logs, memory, vaults, tool outputs, and sandbox files in your own Cloudflare account, Docker deployment, or Node backend.
Self-hosted Claude Tag
Claude Tag is the right product shape: tag Claude where work already happens, give it approved tools, and let it complete tasks with shared context. Open Managed Agents gives you the self-hosted foundation for that pattern: durable sessions, MCP tools, vault-backed credentials, memory stores, sandboxed execution, and BYOK model cards.
The public product surface can still be Slack, because that is where Claude Tag works today. The SEO and product category, though, is self-hosted Claude Tag: an owned runtime behind the same interaction model.
Claude Tag-style channel mention
-> Open Managed Agents Slack publication
-> durable session and event log
-> harness + model card + MCP/tools + sandbox
-> thread reply, memory update, audit trail Run Open Managed Agents on Docker, Node.js, or Cloudflare Workers.
Create a Claude Tag-style agent with a model card, system prompt, memory policy, and approved tools.
Publish that agent into Slack through an Open Managed Agents Slack publication.
Bind MCP servers, GitHub/Linear tools, vault credentials, and sandbox permissions.
Enable the channels where the self-hosted Claude Tag experience should work.
Keep event logs, memory, vaults, tool outputs, and sandbox files in your own Cloudflare account, Docker deployment, or Node backend.
Route model calls through Anthropic, OpenAI, OpenRouter, or an internal gateway while keeping the same agent runtime.
Connect MCP servers, GitHub, Linear, browser, shell, files, and private APIs behind explicit access policies.
Change retrieval, compaction, prompt caching, retry policy, stop conditions, and tool-delivery logic instead of accepting a fixed vendor loop.
Anthropic's Claude Tag is a hosted product. Open Managed Agents gives teams an open-source foundation for building a self-hosted Claude Tag-style experience with their own runtime, model keys, vaults, tools, memory stores, and audit boundary.
It is an open-source runtime foundation, not a finished clone. The core primitives are present: Slack publication, durable sessions, MCP tools, vaults, memory stores, sandboxed work, and recovery. A polished Tag-level admin console is product work on top.
No. You can use Claude through Anthropic, but Open Managed Agents also supports model cards for other providers or a private model gateway.
Self-hosting is useful when channel context, tool calls, repository access, credentials, memory, and audit logs need to stay under your own deployment and compliance boundary.