NanoClaw Lets Slack Users Spawn Persistent AI Agent Teams by Message
NanoClaw lets Slack users create persistent, self-hosted AI agent teams by message, with separate identities, tools, memory and cross-channel access.
Summary
On Aug. 20, 2026, NanoCo introduced a free Slack Marketplace integration for its MIT licensed NanoClaw agent harness, enabling one message to create persistent, independently named bots with generated avatars, roles, memory, instructions, permissions, skills and tools. After cloning NanoClaw and running nanoclaw.sh to build its container, users authorize Connect Slack once; a lead agent then uses Model Context Protocol tools to provision teammates and shared rooms without repeated app configuration. Agents coordinate in channels and Slack Canvas, answer only when tagged, and retain one workspace and memory across separate Slack, WhatsApp and Telegram sessions. Organizations choose Claude Code by default or Codex, OpenCode, local Ollama and other models.
Agents run in separate sandboxes on customers' local machines or cloud VMs and connect through Socket Mode; Slack tokens, data and memory remain local. NanoCo's provisioning and avatar service receives neither messages nor memory, Cohen said. Slack's app approvals apply, and its Home tab can revoke agents or disconnect the workspace. NanoCo charges nothing for the community integration, but users pay model inference and hosting costs. NanoCo calls conversational recursive provisioning a Slack first, although that claim is its own; Claude Tag, ChatGPT Workspace Agents and Agentforce already provide persistent specialists, but create them through administrator or external builders rather than an existing Slack agent.
NanoClaw began Jan. 31, 2026, when former Wix engineer Gavriel Cohen released roughly 500 lines of TypeScript as a container-isolated OpenClaw alternative. Docker Sandboxes added MicroVM isolation in March; NanoClaw 2.0, Vercel's Chat SDK and OneCLI's approval-based credential gateway followed in April. In May, Gavriel and Lazer Cohen formed NanoCo and raised $12 million from Valley Capital Partners, Docker, Vercel, monday.com and others. NanoClaw now claims 250,000 downloads and 30,000 GitHub stars; NanoCo plans paid managed enterprise deployments while keeping it open source.
Positives
- One workspace authorization lets NanoClaw provision additional Slack bots with separate names, generated avatars, identities and tokens.
- Customer-hosted sandboxes keep agent data, memory and Slack tokens on local machines or organization-controlled cloud VMs.
- Slack Canvas, shared rooms and agent tagging support coordinated work while tag-only replies reduce endless bot-to-bot triggering.
- Claude Code, Codex, OpenCode, local Ollama and other model options let organizations balance performance, cost and control.
- 250,000 downloads and 30,000 GitHub stars indicate substantial adoption since NanoClaw's Jan. 31, 2026 release.
- $12 million from Valley Capital Partners, Docker, Vercel, monday.com and others supports NanoCo's managed enterprise expansion.
Risks & concerns
- Users must still clone NanoClaw, run nanoclaw.sh, manage dependencies and credentials, build containers and operate the agent runtime.
- Self-hosting shifts infrastructure reliability, permissions, memory governance, spending controls and agent failure management onto each organization.
- Model inference and hosting remain user expenses even though NanoCo provides the community Slack integration without charge.
- NanoCo's cloud service remains necessary for Slack provisioning and avatar generation, despite receiving no messages or agent memory.
- NanoCo's claimed Slack first is company-supplied, while Claude Tag, ChatGPT Workspace Agents and Agentforce already offer persistent specialized agents.
- Slack app approvals and authorization policies can delay deployment in governed enterprise workspaces.