Skip to content
openma

Claude Tag alternative

Open-source, self-hosted foundation for Claude Tag-style agents

Claude Tag is compelling because it turns Claude into a shared workplace agent with approved tools, scoped context, and async execution. Open Managed Agents gives teams the open-source base to build that pattern themselves, with BYOK model cards, MCP tools, encrypted vaults, memory stores, durable sessions, and sandboxed work under their own deployment boundary.

Star on GitHub Self-hosted Claude Tag

Search intent match

  • Open source: the runtime, integrations, vaults, memory, sandbox, and harness are inspectable and forkable.
  • Self-hosted: run in your Cloudflare account, Docker, or Node backend instead of a closed vendor surface.
  • Tag-style: Slack remains the current user surface, but the category is Claude Tag alternative, not generic bot software.

Claude Tag vs Open Managed Agents

Open Managed Agents should not be positioned as a generic workplace bot. The sharper claim is this: it is an open-source, self-hostable runtime for teams that want the Claude Tag product pattern without giving up deployment, model, tool, memory, and audit control.

Area Claude Tag Open Managed Agents
Source Closed hosted product Apache 2.0 open-source runtime
Hosting Anthropic-hosted for eligible Claude teams Self-host on Cloudflare Workers, Docker, Node.js, or use hosted openma
Primary product shape Claude tagged in Slack with admin-approved tools Claude Tag-style runtime you can productize on your own deployment
Model and keys Claude inside Anthropic's managed service BYOK model cards for Anthropic, OpenAI, OpenRouter, or your own gateway
Tools Admin-approved tools, credentials, and repositories MCP, Slack, GitHub, Linear, shell/browser tools, and custom providers
Memory Productized workspace and channel memory Memory stores under your deployment; Tag-level memory UI is product work
Governance Spend limits, audit, and access bundles in Claude admin Vaults, event logs, quotas, and usage stores exist; admin UX is roadmap

What is covered today

Tag-style channel runtime

Slack publication, mentions, thread replies, and per-channel durable sessions provide the collaboration surface behind Claude Tag.

Open tool and model layer

Use BYOK model cards, MCP servers, GitHub, Linear, browser/shell tools, and private APIs instead of one closed tool surface.

Owned infrastructure

Keep logs, vaults, memory stores, model keys, and sandbox outputs in your own deployment boundary.

Roadmap to Tag-level polish

  1. Claude Tag setup console: workspace pairing, channel enablement, publication binding, and health checks.
  2. Access bundles: inherited workspace/channel policies for tools, credentials, repos, and instructions.
  3. Memory controls: scoped channel memory, admin review/edit/delete flows, and retention policy.
  4. Spend controls: per-channel caps, alerts, blocked-state UX, and usage analytics.
  5. Audit: task history, tool calls, network calls, channel backlinks, and exportable evidence.
  6. Ambient mode: safe background triggers, digesting, stalled-thread detection, and anti-spam rules.

FAQ

Is there an open-source Claude Tag?

Anthropic's Claude Tag is not open source. Open Managed Agents is an open-source foundation for building a Claude Tag-style product with self-hosting, BYOK model cards, MCP tools, vaults, memory stores, and durable sessions.

Is Open Managed Agents a Claude Tag clone?

No. It is the runtime layer under a Claude Tag-style product. The missing layer is a polished Tag admin console for channel enablement, memory controls, access bundles, spend limits, and audit dashboards.

Can Open Managed Agents cover the Claude Tag feature shape?

It covers much of the runtime shape: Slack publication, per-channel sessions, MCP/private tools, vault credentials, memory primitives, sandboxed work, and recovery. Tag-level governance and UX are roadmap work.

Why use Open Managed Agents instead of Claude Tag?

Use it when you need open source, self-hosting, model-key ownership, private tools, custom harness behavior, and control over event logs, credentials, memory, and sandbox output.