By admin 
The promise of integrating AI agents into enterprise Slack workspaces is a tantalizing prospect for many organizations, offering the allure of enhanced productivity and streamlined workflows. However, as VentureBeat has personally experienced, the practical implementation often proves more complex and cumbersome than initially anticipated. Addressing this challenge head-on, NanoCo., the innovative company behind the open-source, enterprise-friendly autonomous AI agent harness NanoClaw, is poised to transform this landscape. NanoClaw, a more sandboxed and low-code iteration of OpenClaw, aims to make AI agent deployment as effortless as sending a Slack message. Their groundbreaking new NanoClaw Slack integration empowers human users to orchestrate entire teams of specialized AI agents, complete with distinct skills, custom workflows, and even personalized avatars, all initiated through a simple Slack prompt.
"In the next 12 to 18 months, everyone on a team will be a manager of agents," declared NanoCo CEO and co-founder Gavriel Cohen in an exclusive interview with VentureBeat. This bold prediction underscores the company’s vision for a future where AI agents are not merely tools but integral, collaborative members of the workforce.
Further enhancing this vision, NanoClaw agents are designed to interact seamlessly within Slack channels and shared Slack Canvases. Their utility extends beyond the confines of Slack, allowing human colleagues to communicate with them on other platforms like Telegram or WhatsApp, mirroring the familiar experience of messaging human team members across different applications. "I think this is agents arriving natively in Slack for the first time," Cohen emphasized. "In the past, you had to do all these weird things to try to have multiple different agents behind the scenes using the same bot, and now every agent gets its own identity in Slack – its own avatar, its own face, its own name. You can tag them. They can tag each other."
A critical differentiator for enterprise teams lies in NanoClaw’s emphasis on persistence and separation. Unlike ephemeral chatbots that disappear after a single task, each NanoClaw agent can be assigned a specific role, memory context, detailed instructions, and granular permissions. This creates a robust structure akin to a miniature digital department rather than a rudimentary chatbot controlled by an extended prompt. Echoing the ethos of the original open-source NanoClaw released in January 2026, developers and enterprises retain the flexibility to select their preferred underlying large language model (LLM) to power their NanoClaw agents, optimizing for performance, cost-effectiveness, or a tailored combination of factors.
From a Single NanoClaw Slack Agent to a Specialized, Collaborative Team
The initial setup for NanoClaw, while requiring some technical engagement, has been significantly streamlined. The current process involves cloning the project and executing the nanoclaw.sh installer. This installer guides users through dependency management, credential configuration, agent container building, and the pairing of the first messaging channel. NanoClaw’s website assures users that the installer can take them "from a fresh machine to a named agent you can message," with Slack being one of the prominently supported channels.
Cohen elaborated on the simplified Slack-specific workflow, highlighting its substantial advantage over traditional Slack bot development. Previously, users would have to navigate Slack’s complex administrative and developer interfaces, painstakingly create an application, collect secrets, API keys, and tokens, and then meticulously transfer these credentials to the bot’s running environment. The new NanoClaw integration radically alters this paradigm. Instead, it presents a straightforward "Connect Slack" option. Users simply name their agent, authenticate their identity, select the "Add to Slack" option within NanoClaw, and proceed through Slack’s standard installation and authorization process. Upon successful authorization, the first agent materializes within Slack, ready to engage with the user.
A pivotal distinction of this integration is the largely one-time workspace connection. Slack’s Marketplace listing explicitly states that users "connect a workspace once," after which NanoClaw can provision each subsequent agent as a distinct Slack bot, complete with its own unique name, a dynamically generated avatar, and a dedicated identity. These agents operate on the customer’s own infrastructure and connect to Slack via Socket Mode. NanoCo. asserts that it does not store agents’ Slack tokens; according to the Marketplace listing, these tokens remain securely on the user’s machine.
Crucially, Slack’s established administrative controls remain firmly in place. Organizations can seamlessly apply their existing app-approval policies to the NanoClaw integration. Furthermore, NanoClaw’s Marketplace listing details that the app’s Home tab provides a clear overview of all provisioned agents within a workspace, allowing users to revoke individual agents or disconnect the entire workspace at their discretion. The outcome is less a one-click replacement for NanoClaw’s underlying infrastructure and more of a robust, one-time bridge between that infrastructure and Slack. While users retain ownership and operational control of the agent runtime, once the bridge is authorized, the agents themselves can autonomously create and coordinate additional Slack-native colleagues without requiring the user to re-enter the manual app configuration process for each new addition.
Behind the scenes, Cohen explained, a lead agent is equipped with a Model Context Protocol (MCP) tool. This powerful tool enables the creation of new agents, defining their specific instructions, personas, skill sets, and available tools. Another integrated tool facilitates their placement into shared communication rooms. The agents are pre-configured to work natively with Slack Canvas and can engage in fluid communication with every human user within a Slack Channel, as well as with each other.
The interaction design is intentionally intuitive. Instead of requiring users to open a separate agent builder each time a new role is needed, Cohen described how users can simply instruct their existing agent about the type of colleague or team they require. "Your agent in Slack, you can say, ‘Create me another agent to handle my code reviews. Create another agent to review the contributor articles. Create a team of agents that reviews contributor articles from different perspectives.’ And then your agent can create new agents, and they just pop up in the sidebar and send you messages." This capability allows a developer, for instance, to request the creation of a product manager, an architect, an implementation agent, a code reviewer, and a testing agent, each with a distinct toolset, and then have them efficiently hand off work amongst themselves. Cohen cited the example of a testing agent having access to a dedicated testing environment, while a review agent possesses specialized code-review skills, and a product agent monitors user feedback, demonstrating the power of this task-specific delegation.
Cohen firmly believes this division of labor transcends mere cosmetic role-playing. "There are advantages in terms of giving each one specific skills, instructions, and tools for different tasks," he stated. "I can have, let’s say, a code review agent, a code testing agent, a code writing agent, and I can have them in a loop." If an implementation agent encounters ambiguity, Cohen added, it can tag the product or architecture agent for clarification, rather than burdening a single general-purpose model with every responsibility and tool within the same context.
Agents Collaborate with Humans on a Shared Slack Canvas
A provided demonstration screenshot vividly illustrates this collaborative model applied to marketing operations. A lead agent named Nano orchestrates Atlas for strategy, Sage for content creation, Echo for social media engagement, Scout for outreach initiatives, and Compass for SEO and analytics. The agents introduce themselves within the same Slack conversation and immediately begin coordinating their efforts. Atlas, for example, notes its action of adding an item to Canvas, ensuring the task remains trackable and visible.
Users are not required to specify every minute detail upfront. Cohen explained that individuals can provide the lead agent with precise review procedures, priorities, and essential tools, or delegate a greater degree of configuration to the agent based on its existing context and accumulated memory. The design also actively seeks to mitigate a common multi-agent failure mode: bots endlessly triggering each other. NanoCo. asserts that agents respond only when explicitly tagged, and comments made on work within Canvas can be intelligently routed back to the agent responsible for that particular task.
The model’s potential extends beyond teams of task-specific bots created by a single individual. Cohen envisioned a workplace scenario where each employee possesses a persistent agent capable of inter-agent communication, governed by human-defined policies. "Each person having their own agent means that I could have my agent and you have your agent in Slack, and your agent can ask my agent questions," he posited. "Maybe I’m out of the office for the day. Your agent can ping my agent and ask a question about availability, and I can set some policies about whether my agent can answer or if I need to give approval." This paradigm shifts the concept closer to organizational delegation, where some agents specialize by function, while others effectively represent individual employees and the unique context they have cultivated. Cohen further noted that these agents can be equipped with browser and internet access, extensive memory capabilities, coding functionalities, and a diverse array of other tools. Newly created agents arrive with built-in support for Canvas collaboration, seamless agent-to-agent communication, and the ability to spawn even more agents.
Slack Opens the Door for Expanded Third-Party Agent Integration
This underlying shift in Slack’s architecture is more profound than a single integration. In April, Slack, a Salesforce product, announced the capability to directly integrate external AI agents into its messaging platform. Initially highlighting Vercel and Lovable, the company indicated that these integrations would become available in late May. Slack’s announcement detailed a deployment mechanism that automates OAuth, manifest configuration, and environment setup, enabling externally developed agents to be brought into a workspace without requiring a complete rebuild specifically for Slack.
Salesforce’s newly launched Slack Code page now prominently features NanoClaw alongside Lovable, Hyperagent, Superhuman, n8n, Vercel, ChatGPT, LangChain, Runlayer, and Skydive. The platform states that the "Add to Slack" functionality can seamlessly bring agents from these platforms into Slack with their own distinct identities in just a few clicks.
The Slack ecosystem is already a vibrant hub for AI assistants. OpenAI, for instance, permits the deployment of ChatGPT workspace agents into Slack channels, where they can field questions, execute tasks through connected systems, and generate output files. Slack also supports Claude and custom Agentforce agents. Consequently, NanoClaw’s unique selling proposition is not merely the presence of "AI in Slack." Its true innovation lies in the capability for an already operational agent to dynamically create additional, independently addressable teammates directly from within the ongoing conversation itself. NanoCo. asserts this is a first for Slack, a claim that marks the company’s distinctive market entry.
"Add to Slack means one message can spin up a full team of NanoClaw agents, working right alongside people in Slack," stated Josh Milas, director of product management at Slack, in the official announcement.
NanoClaw Differentiates Itself from Claude Tag, ChatGPT Agents, and Agentforce in Slack
NanoClaw is not operating in a vacuum as it endeavors to transform AI from a mere sidebar chatbot into a persistent, integrated Slack colleague. Anthropic’s Claude Tag, which began its beta rollout to Claude Team and Enterprise customers in June, presents the closest conceptual parallel. Administrators can grant @Claude access to selected channels, tools, data sources, and codebases, empowering all channel participants to delegate tasks by simply tagging it. Claude possesses the ability to recall relevant information from the channels it inhabits, can operate asynchronously over extended periods, and, when administrators enable its "ambient" behavior, can proactively flag information or revive unresolved work without requiring an explicit prompt. Anthropic also allows for separate Claude identities to be scoped to different use cases, ensuring that, for example, a sales-focused Claude does not share its memories or tools with an engineering-focused Claude.
The fundamental divergence lies in the method of provisioning and organizing these digital coworkers. Claude Tag’s documented workflow is administrator-centric: administrators pair Claude with Slack, meticulously determine which channels, tools, and information each Claude identity can access, set spending limits, and then expose these identities to employees. Within a given channel, Anthropic describes "one Claude that interacts with everyone." Its public documentation does not detail a scenario where an end-user can instruct that Claude to generate multiple new, independently named Slack bots on demand.
NanoClaw’s model operates on an almost inverted principle. Following an organization’s one-time connection of its NanoClaw installation to Slack, NanoClaw claims that an existing agent can autonomously provision additional agents through a conversational request. Each newly created worker is assigned its own unique Slack bot identity, name, a generated avatar, and a token, and operates back on the customer’s infrastructure.
OpenAI’s ChatGPT Workspace Agents occupy another position on this spectrum. Business, Edu, and Enterprise customers can construct reusable agents within ChatGPT, furnish them with instructions, models, files, applications, custom MCP connections, and schedules, and subsequently attach these agents to Slack channels. Builders assign each agent a unique Slack handle and can configure it to respond solely when mentioned or to automatically react to relevant messages within a channel. However, the agent construction process primarily occurs through ChatGPT’s dedicated agent builder. OpenAI’s setup documentation guides users to create the agent first and then add Slack as a channel. Internally, the Slack handles leverage Slack user groups managed by the ChatGPT Agents app, rather than NanoClaw’s approach where each provisioned agent functions as a separate Slack bot.
Salesforce’s Agentforce similarly empowers organizations to create multiple specialized agents that employees can interact with via direct messages or @mentions within Slack. It arguably offers the most conventional enterprise administration model among the group. Companies build agents within the Agentforce Builder, often commencing with Slack-specific templates tailored for roles such as customer insights, employee assistance, or onboarding. They can also incorporate subagents and actions that enable information retrieval, Canvas creation, or other workflow-related tasks. Once configured and activated within Salesforce, administrators introduce these agents into Slack for employee utilization. This makes Agentforce particularly potent for organizations that already centralize identity, data, and workflows on the Salesforce platform. However, it again places agent creation prior to deployment, rather than making the creation process itself an interactive capability of an existing Slack agent during a conversation.
This critical distinction clarifies NanoClaw’s unique contribution to an increasingly populated market. Slack itself now provides an Agent Kit for developers and a deployment standard for agents built on external platforms, automating essential components such as OAuth, manifests, and environment configuration. Claude Tag, ChatGPT Workspace Agents, and Agentforce collectively demonstrate that persistent, specialized AI teammates within Slack are no longer novel in isolation. NanoClaw’s more unconventional proposition is recursive provisioning: Slack evolves from being merely the arena where workers invoke agents to a dynamic environment where an existing agent can assemble additional named agents, assign them distinct roles, and integrate them into a channel as a functioning team.
Each of these approaches comes with its own set of trade-offs. Claude Tag benefits from Anthropic-managed models and centralized administrative controls, including channel-specific permissions, comprehensive audit logs, and token-spending limits, while also offering proactive "ambient" behavior that NanoClaw’s provided materials do not explicitly claim in the same manner. ChatGPT Workspace Agents provide a managed agent builder, scheduling capabilities, application integrations, and organization-level publishing and access controls. Agentforce tightly integrates agents with Salesforce permissions, enterprise data, and predefined business actions.
NanoClaw, conversely, emphasizes self-hosting, open-source modifiability, and distinct agent identities, thereby shifting a greater degree of control – and operational responsibility – to the organization implementing it. The result is less a direct replacement for these existing systems and more a distinct answer to the emerging question: whether enterprises desire a limited number of centrally configured AI assistants or an environment where employees and existing agents can continuously generate specialized digital colleagues as new work requirements arise.
The Genesis of NanoClaw
NanoClaw’s journey began far from the realm of enterprise collaboration. Gavriel Cohen, a former engineer at Wix, initially launched it under the MIT License on January 31, 2026. His vision was to create a deliberately lightweight, security-focused alternative to OpenClaw. The original proposition centered on the idea that a personal agent, with access to messages, files, and tools, should operate within an OS-isolated container rather than directly on the host system. The orchestration layer was designed to remain small enough for a developer or security team to fully comprehend – initially comprising approximately 500 lines of TypeScript and a design focused on container isolation and a minimal single-process architecture.
The project subsequently evolved with a clear trajectory toward enterprise infrastructure. In March, NanoClaw partnered with Docker to facilitate the execution of agents within Docker Sandboxes, leveraging robust MicroVM-backed isolation for workloads that might involve package installation, file modification, and process launching. By April, NanoClaw 2.0 introduced Vercel’s Chat SDK and OneCLI’s credential gateway, empowering organizations to define policies around sensitive actions and mandate human approval before credentials are injected for protected requests.
By May, Cohen and his brother, Lazer Cohen, had established NanoCo. around the project, successfully raising a $12 million seed round led by Valley Capital Partners, with participation from Docker, Vercel, monday.com, and other prominent entities. The company’s commercial strategy involves maintaining NanoClaw as open-source while offering managed, organization-wide deployments and "professional assistant" infrastructure to enterprises. NanoCo. now reports that NanoClaw has surpassed 250,000 downloads and garnered over 30,000 GitHub stars.
This open-source foundation remains a cornerstone of Cohen’s strategic vision as NanoClaw penetrates deeper into workplace infrastructure. "You’re really able to now integrate an open-source agent into Slack that you fully control," he stated. "You can change all those configurations. Plus, you can fork NanoClaw and completely rewrite or change behaviors – create your own memory system, your own coding harness, agent harness. Whatever you want to do, you can do. Total freedom."
Persistent Agents, but Infrastructure Remains Under User Control
Cohen reiterated that NanoClaw maintains its self-hosted architecture: organizations can deploy it on a local machine or their own cloud virtual machine, with agent data stored locally. The same agent can seamlessly appear across Slack, WhatsApp, or Telegram while retaining its unique memory, workspace, and tools, although each messaging surface operates as a distinct session. NanoClaw is designed to pull recent context across these sessions, enabling the agent to maintain continuity without merging every chat history into a single, monolithic stream. NanoClaw’s documentation similarly outlines a multi-channel architecture where a single agent can preserve one workspace and memory while managing separate per-channel sessions.
"This is all self-hosted," Cohen emphasized. "You’d be running this on your computer or on your virtual machine in the cloud, and that data is stored on your computer or on your [virtual machine] VM. This could be an open-source model running on your Mac Mini, and your data isn’t going anywhere besides your Mac Mini and then into Slack." The cross-channel continuity is also intended to foster the perception of an agent as a persistent colleague, accessible through Slack, rather than merely a Slack-specific bot. Cohen elaborated that the same agent could exist across Telegram, WhatsApp, and Slack, with access to the identical memory, files, and tools. While the conversations remain distinct sessions, they share a common workspace and persistent context, allowing the agent to transfer knowledge seamlessly between different surfaces.
This architecture becomes particularly significant when an organization begins creating a multitude of agents. Cohen explained that one agent can access its own sessions across channels, but not another agent’s private sessions by default. NanoClaw’s current documentation similarly describes agents operating within their own sandboxes and configurable model providers, with Claude Code serving as the default, and Codex, OpenCode, and local Ollama models available as alternative options.
There is one cloud-dependent component for the new Slack integration: Cohen mentioned that NanoCo. operates a small service responsible for handling Slack provisioning requests and avatar generation. He clarified that this service does not access users’ messages or agent memory.
Continued Commitment to Open Source
According to Cohen, NanoCo. is not imposing charges for this community-focused Slack capability, absorbing the costs associated with the provisioning service and avatar generation. Users will still incur their own model inference and hosting expenses, meaning a deployed agent team is not entirely cost-free in practice. NanoCo. states that the integration is accessible through the Slack Marketplace, subject to standard workspace app approval and governance procedures. Slack itself mandates that workspace owners and administrators can require apps to be approved prior to installation.
Cohen framed this decision as an integral part of NanoCo.’s broader open-source strategy rather than a standalone monetization initiative. "We’re not making any money off this one. This one is for the community, really," he asserted. "We know that in the long run, that’s going to benefit NanoCo as a company. As NanoCo grows and builds out capabilities, those go back to the open source. I think that’s the new model of open source, where we’re not trying to monetize every bit of value we bring to the community."
Whether companies achieve this level of integration rapidly will likely depend less on the ease of agent creation and more on the capacity of IT teams to govern their permissions, memory usage, spending, and potential failure modes at a commensurate pace. NanoClaw is strategically betting that the next significant challenge for enterprises will be effectively managing the burgeoning digital workforce that emerges once the initial barrier to creation is overcome.
