Meet Jarvis, Your AI Workforce

Agently gives you Jarvis, a single AI agent that runs your workspace. You don't pick a bot for the job from a menu. You tell Jarvis what you need in plain language, and it works the problem: it reads the request, pulls context from your Brain, acts step by step through your connected apps, and spins up focused sub-agents when a job is large enough to split. It all happens in one conversation, in the Command Center.

Jarvis is powered by Anthropic's Claude (Opus 4.8). Unlike a general-purpose chatbot, it doesn't just describe what to do, it does it: sends the email, creates the calendar event, updates the board, publishes the page, and shows you a citation for every fact it used.

What makes this different from a chatbot:

• One brain that compounds. Everything your workspace learns (uploads, connected apps, past work) becomes a living company knowledge graph Jarvis reasons over. It gets smarter the more you use it.
• One agent that orchestrates many. Instead of you juggling a menu of narrow bots, a single orchestrator plans the work and spins up focused sub-agents for the heavy parts.
• Skills that teach it your playbooks. Package a repeatable workflow once, and Jarvis runs it your way every time.

Upgrading from the old agents? Earlier versions of Agently shipped a roster of named specialists, Apex, Nova, Echo, Pulse, Lens. Those are gone. Everything they did now lives in Jarvis, which covers sales, operations, marketing, customer success, and research from one conversation, and delegates to specialist sub-agents behind the scenes when it helps.

How Jarvis Works: The Agentic Loop

Jarvis doesn't answer in a single shot. It runs an agentic loop, a plan-act-observe cycle that repeats until the job is done:

1. Understand. Jarvis reads your message and searches your Brain for relevant company context.
2. Act. It picks a tool (search the web, draft in Gmail, create a task) calls it, and waits for the result.
3. Observe and decide. It reads what came back and decides the next step: use another tool, delegate a sub-task, or write the final answer.
4. Repeat until the work is complete, then return the finished output.

This is why a single request like "research this prospect, draft the intro email, and schedule a follow-up" can complete end to end in one turn, Jarvis chains the steps itself. It streams its progress live, so you watch the work happen rather than staring at a spinner. Every run is bounded by a safety limit and your credit budget, so it can't loop forever.

Grounded in Your Company Knowledge

What makes Jarvis's answers yours and not generic is the Brain, a temporal knowledge graph of your company. When Jarvis searches it, the Brain returns ranked facts, each with a validity window (when the fact was true) and a citation back to the source.

Two things follow from that:

• Answers are cited. When Jarvis tells you something about your business, it can show you exactly which document, message, or record it came from.
• Answers respect time. Because facts carry validity windows, Jarvis can reason about what's true now versus what changed, it isn't fooled by a stale doc that a plain keyword search would surface.

That's the real innovation: your company's knowledge stored the way it actually behaves (connected facts that change over time) so Jarvis reasons over what's true now, with receipts. See Brain for how content gets in and how retrieval works.

Sub-Agents: How Jarvis Delegates

When a request contains a substantial, self-contained chunk of work ("research these ten competitors," "write the full launch announcement") Jarvis can delegate it to a sub-agent instead of doing everything in the main thread.

A sub-agent is a fresh worker with:

• Its own clean context window, focused only on the delegated task, so the main conversation stays fast and uncluttered.
• Its own slice of the run's budget, so delegation is cost-controlled and predictable.
• A specialist focus. Under the hood Jarvis routes to the right flavor for the job (content, sales, research, or operations) each with a curated toolset.

Sub-agents run one level deep (a sub-agent can't spawn its own sub-agents), return a summary of what they did, and hand the result back to Jarvis. You never manage any of this: Jarvis decides when to delegate, there's no sub-agent to summon or pick. It just works like a lead who parcels out the heavy lifting and reports back, all in one thread.

What Jarvis Can Actually Do

Jarvis's real power is its tools, the concrete actions it can take. Here's the full set, grouped by what you'd use it for.

Know your business

• Brain / Knowledge Base, searches your company knowledge graph for cited, time-aware answers, and saves new snippets and web pages back to it. See Brain.

Create and publish

• Pages (creates and edits documents, presentations, spreadsheets, and interactive HTML artifacts) and can publish a page as a live public website at a yourslug.agently.dev address. See Pages.
• Content creation, a dedicated writing pass for polished deliverables (social posts, LinkedIn posts, blog posts, emails, articles, ad copy) with tone and audience controls.
• Image generation, generates images from a text prompt and returns a shareable URL. (Generating an image spends extra credits.)
• Voice samples, pulls your own recent posts from a connected account so Jarvis can write in your actual voice, not a generic one.

Organize work

• Spaces, creates and manages Kanban boards, folders, columns, and tasks, including bulk create/update. See Spaces.
• Scheduler, sets up one-time or recurring runs, including scheduled posts to X/Twitter and LinkedIn. See Calendar.
• Support tickets, files and tracks tickets for issues, bugs, and feature requests.

Act in your connected apps

Through your integrations, Jarvis reads from and writes to your real accounts:

Every one of these actions is written to an audit log, so there's always a record of exactly what Jarvis did on your behalf.

Research the open web

• Web search and web fetch, search the web and read specific URLs for live, up-to-the-minute information.
• Code execution, runs code to calculate, analyze data, and process information precisely.

Staying in Control: Actions & Approvals

For everyday chat, Jarvis acts directly, you're in the conversation, so you see and direct each step. For scheduled or autonomous runs (work Jarvis does when you're not watching), Agently adds a governance layer:

• Approval queue. You can require that certain actions (say, sending an external email) be proposed rather than executed. Proposals land in your Inbox as Decisions for you to approve or reject. Approve one and Jarvis replays the exact action.
• Policy per action. Each app action can be set to run automatically, require approval, or be blocked outright. Anything not explicitly allowed fails safe (blocked).
• A kill switch and full audit trail cover autonomous work end to end.

See Inbox & Notifications for the approvals surface and Security & Privacy for how connected accounts are secured.

What Runs Support

Alongside Jarvis, Agently runs a lightweight Support Assistant on a faster, cheaper Claude model (Haiku) for the in-app help widget. It answers product questions from the Agently knowledge base and can file a support ticket, it's not part of your workspace's working agent, just the help you get inside the app.

Example Things to Ask Jarvis

Because Jarvis covers every role, you can hand it work across your whole business in one place:

• "Research the top 10 SaaS companies in HR tech and put a prospect brief for each into a new page."
• "Draft a 3-email cold outreach sequence for CTOs at mid-market fintechs, use our positioning from the Brain."
• "Check my calendar this week, find a 30-minute slot for a team sync, and send the invite."
• "Plan our Q2 launch: set up a board in Spaces with tasks, owners, and deadlines."
• "Write a LinkedIn post announcing our new feature in my voice, and schedule it for Tuesday at 9am."
• "Do a competitive analysis of our top 5 competitors, compare pricing, features, and positioning, with sources."
• "Create a support ticket for the billing issue Acme reported, then draft a reply."

Not sure how to phrase something? Just describe the outcome you want. See Chatting with Agents for how to get great results.

Next Steps

• Getting Started, Set up your workspace and send Jarvis its first task
• Core Concepts, The building blocks behind the platform
• Chatting with Agents, Write prompts that get great results
• Use Cases & Playbooks, Real workflows you can copy