# Agentic Workspace- specialist agents that operate NexoFlow Source: https://nexoflow.ai/workspace Eight agents- an orchestrator plus SEO, Content, Publishing, Analytics, Social, Research and Branding- holding 69 tools, with 17 high-impact tools gated behind an explicit approval click. ## What it is The Agentic Workspace is not a chat box beside the product. It is the product, driven by a team of specialist agents that create the project, scan the website, write the plan, generate the images, and put the posts on the calendar. Most product chatbots answer questions about the product. This one runs it: “Set up a project for this client and draft next month” is a request it carries out end to end, because the agents hold the same tools the dashboard buttons call. 8 specialist agents · 4 modes · 69 tools · every publish behind an approval ## One conversation, a team behind it 1. Say what you want- Plain language. No command syntax, no prompt engineering. Pick a mode if you like- Plan, Work, Review, or Automations- or just start typing. 2. An orchestrator hands it off- Seven specialists sit behind it: SEO, Content, Publishing, Analytics, Social, Research, and Branding. You watch each step run, with its own progress and logs. 3. You approve what matters- Drafts, plans and previews land inline. Anything public or destructive stops and waits for an explicit Approve click before it runs. ## Things people hand it Each of these maps onto shipped tools, not a roadmap promise. - “Onboard this new client”- Create the project, scan their website, autofill the brand profile from it, and connect WordPress- in one conversation. - “Plan next month for me”- Build blog and social plans, adjust the ones that miss, and schedule what survives review. - “Learn our documentation”- Add URLs, PDFs, and Drive folders as knowledge sources, then check what finished processing and what failed. - “We need images for these”- Generate media library images and featured images for specific posts, without leaving the thread. ## Agents with real reach, kept in bounds The workspace runs inside your account and your permissions- it is not a separate system with its own access. - Publishing waits for a click- Seventeen high-impact tools- publishing, scheduling, deleting, inviting, replying to reviews- are gated. The agent proposes; you press Approve. - You watch it work- Each action shows its agent, its tool, its progress and its logs while it runs, then the result it produced. No silent background edits. - Scoped to your projects- Agents act within the projects your account can reach. They cannot wander into another organization's data. - Your plan still applies- Limits, credits, and tier gating are enforced on the same server paths as the rest of the app. The workspace is not a bypass. - Drafts before publishes- Default output is drafts and plans in the normal dashboard views, reviewable the same way your own work is. - Same platform over MCP- Prefer your own AI client? The MCP server exposes a deliberately narrower slice of this to Claude, Cursor, and anything else that speaks MCP. The workspace, or your own AI client: The Agentic Workspace is the fuller surface- it sets up projects, manages knowledge, generates media, and shows you every step. The MCP server is the narrower one, built for handing an outside AI client scoped access with a cancel window on every write. Same underlying platform, two different trust models, and you pick per situation. ## A sample of what the agents can call- Reads and diagnoses - get_project_status- Where a project stands right now - get_social_connections- Which channels are connected and healthy - diagnose_wordpress_connection- Why a WordPress publish is failing - get_knowledge_status- What finished processing, what failed - list_media_library- Images available to reuse - get_org_team_usage- Output and spend per member ## A sample of what the agents can call- Creates and publishes - create_project- Stand up a new brand workspace - scan_website- Crawl a site to learn the brand - create_blog_schedule_plans- Draft a month of website content- approval required - create_social_media_plans- Draft per-channel social posts- approval required - generate_media_library_images- Make on-brand images - publish_website_post- Put a post live- approval required ## Common questions Is this the old AI assistant? No- it replaced it. NexoFlow used to have a chat assistant sitting beside the dashboard; the Agentic Workspace is a workspace in its own right, with specialist agents, four working modes, live action tracking and an approval gate. The old chat route now opens the workspace. How is this different from pasting a prompt into ChatGPT? A general chatbot can write you a draft. The workspace can create the project, learn the brand from the website, write the plan into your calendar, generate the images, and schedule the posts- because its agents hold the actual tools your dashboard uses. Can it publish something by accident? No. Seventeen high-impact tools- publishing, scheduling, bulk date shifts, deletes, team invites, Google review replies- cannot run until you click Approve. That gate is also a defence against a scraped page or document trying to talk the agent into something. What are the agents, exactly? An Orchestrator that coordinates, plus SEO, Content, Publishing, Analytics, Social, Research, and Branding specialists. You see which one is handling each step as it runs. Does it know about my brand? Yes, when you have given it something to learn from. It reads the project's brand profile and knowledge sources, and it can build those itself by scanning your website or ingesting documents you point it at. How many tools does it have? Sixty-nine, spanning projects, content plans, knowledge sources, business profile, media generation, team management, connections, and publishing. Can I use my own AI client instead? Yes- that's what the MCP server is for. Claude, Cursor, and any MCP client can connect with scoped tokens. It exposes a narrower, more heavily guarded set of operations than the workspace, with a cancel window on every write. Is my data used to train a model? No. Your content and brand data are used to do the work you asked for. See the Security page for how data is stored and handled.