NanoCo’s NanoClaw now offers a Slack integration that allows users to spawn entire teams of AI agents—with individual roles, permissions, memory contexts and generated avatars—directly from a Slack prompt. The company says the agents remain self‑hosted on the customer’s infrastructure while appearing in Slack as distinct bot identities.
What changed and why it matters
Instead of the traditional multi‑step Slack bot setup, NanoClaw provides a Connect Slack flow: authenticate, choose Add to Slack, and after a one‑time workspace connection the first agent appears in Slack. After that initial authorization, existing NanoClaw agents can provision additional agents on demand; each new worker receives a Slack bot identity, name, and avatar.
Crucially, these are persistent workers rather than ephemeral subagents. Each agent can be assigned its own instructions, memory and tools, producing an architecture closer to a small digital department than to a single chatbot extended by a long prompt.
Installation and technical flow
For a fresh installation, the NanoClaw process begins by cloning the project and running nanoclaw.sh, which guides users through dependencies, credentials, building the agent container and pairing a first messaging channel. Slack is listed as a supported channel; NanoClaw’s Marketplace listing notes that a workspace is connected once and thereafter the system can provision multiple agents as separate Slack bots.
Agents run on the customer’s infrastructure and connect to Slack over Socket Mode. NanoCo says it does not store users’ Slack tokens; according to the Marketplace listing the tokens remain on the user’s machine. NanoCo does operate a small service that handles Slack provisioning requests and avatar generation, which the company says does not receive users’ messages or agent memory.
Interaction and collaboration
NanoClaw agents can work together in channels and shared Slack Canvases, and they can be messaged from other platforms such as Telegram or WhatsApp while retaining the same workspace memory and tools (each platform keeps a separate session). The interaction model is intentionally simple: a lead agent can be told to create additional agents for specific tasks, and those new agents will appear in Slack’s sidebar and begin sending messages. Agents reply only when tagged to avoid bots endlessly triggering one another; Canvas comments can be routed back to the responsible agent.
Gavriel Cohen, NanoCo’s CEO and co‑founder, told VentureBeat that “in the next 12 to 18 months, everyone on a team will be a manager of agents.” The product supports scenarios where one developer asks an agent to create a product manager, architect, implementation agent, code reviewer and testing agent, each with different toolsets and responsibilities.
Examples and use cases
A supplied demo shows a marketing scenario where a lead agent (Nano) creates Atlas for strategy, Sage for content, Echo for social, Scout for outreach and Compass for SEO and analytics. The agents introduce themselves in a Slack conversation and coordinate work, with Atlas adding an item to Canvas so tasks aren’t lost. Users can either provide exact procedures and priorities up front or let the lead agent use existing context and memory to configure new agents.
How NanoClaw differs from other Slack agent approaches
Slack has opened a broader mechanism to add external agents; Slack’s Code page lists NanoClaw alongside Lovable, Hyperagent, Superhuman, n8n, Vercel, ChatGPT, LangChain, Runlayer and Skydive. Still, persistent, specialized AI teammates in Slack are not unique by themselves. The distinction NanoCo highlights is that an already‑running agent can spawn additional, independently addressable teammates from inside a conversation.
Comparisons from the article:
- Anthropic Claude Tag: admin‑led provisioning and channel scoping, with one Claude interacting with a channel’s participants and separate Claude identities administratively bound to use cases.
- OpenAI ChatGPT Workspace Agents: agents are built in ChatGPT first and then attached to Slack channels; Slack handles for those agents rely on ChatGPT Agents app user groups rather than each agent being a separate Slack bot.
- Salesforce Agentforce: agents are created in Agentforce Builder and then deployed into Slack under Salesforce admin control, which places agent creation before deployment.
NanoClaw’s approach is recursive provisioning: after a one‑time connection, a NanoClaw agent can create new agents from a conversational request, and each new worker runs on the customer’s infrastructure with its own Slack bot identity.
Background and development
NanoClaw was released under the MIT License on January 31, 2026, as a compact, security‑focused alternative to OpenClaw. The original design emphasized container isolation and a minimal orchestration layer of roughly 500 lines of TypeScript. The project moved toward enterprise infrastructure: in March it partnered with Docker to run agents inside Docker Sandboxes; in April NanoClaw 2.0 added Vercel’s Chat SDK and OneCLI’s credential gateway. In May Gavriel and Lazer Cohen formed NanoCo and raised a $12 million seed round led by Valley Capital Partners with participation from Docker, Vercel, monday.com and others. NanoCo says NanoClaw has exceeded 250,000 downloads and 30,000 GitHub stars.
NanoCo’s commercial strategy is to keep NanoClaw open source while selling managed, organization‑wide deployments and professional assistant infrastructure to enterprises. Cohen said the Slack capability is being offered to the community without direct charge and that NanoCo is absorbing provisioning and avatar generation costs, although users still pay for model inference and hosting.
Operational trade‑offs and governance
NanoClaw emphasizes self‑hosting, open‑source customization and separate agent identities, which gives organizations control and flexibility (for example, using local models or building custom memory systems) but also places more operational responsibility on IT teams. As Cohen noted, the key challenge will become governance: managing permissions, memory, spending limits and failure modes as organizations create many persistent agents.
Conclusion
NanoClaw’s Slack integration adds a new model to the growing set of ways enterprises bring persistent AI coworkers into collaboration tools: rather than only pre‑building a small set of centrally configured assistants, NanoClaw enables employees and agents themselves to spawn specialized digital colleagues on demand from within Slack conversations. The approach shifts power to organizations running their own infrastructure while raising questions about how IT teams will govern and control the resulting population of digital coworkers.
