Skip to content
Onplana Docs

Set up Onplana Business with AI agents

All plansOwner or Admin

This guide takes a workspace from no agents to an AI agent doing assigned work under your rules, in five steps you can finish in an afternoon. Each step links to the article that owns the detail, so this page stays short.

The short version: steps 1 to 4 work on every plan, including Free. Step 5 shows what the Business plan adds to an agent setup, so you can tell whether your team needs it.

Before you start: be the workspace Owner or an Admin (the agent controls in steps 2 and 3 are Owner and Admin by default), and have one project with a few real tasks in it, so the agent has something to work on.

1. Connect an assistant (about 15 minutes)

Section titled “1. Connect an assistant (about 15 minutes)”

Pick the assistant your team already uses. Every route below ends in the same place: the assistant reaches your workspace through Onplana’s MCP server, with your permissions.

Assistant How it connects Step-by-step
Claude (claude.ai, Claude Desktop) Anthropic’s Connectors Directory, one browser sign-in Connect from the Claude Connectors Directory
ChatGPT OpenAI’s directory, one browser sign-in Connect Onplana to ChatGPT
Microsoft 365 Copilot The Onplana app for Microsoft Teams, which a Microsoft 365 admin deploys once Deploy Onplana for Microsoft Teams
Gemini Enterprise A custom MCP server in Google Cloud. Early access: Onplana issues the client details on request Connect Onplana to Gemini Enterprise
Claude Code, Cursor, Codex and other MCP clients Connect Agent in Onplana, with a browser sign-in or a token Connect an external agent

Check it works: ask the assistant “What is overdue in my Onplana projects?” A working connection answers with your real tasks, and the connection appears under Settings → Agents.

What the first connection costs, on every plan

Section titled “What the first connection costs, on every plan”
  • Connecting is on every plan, Free included. There is no plan gate on the MCP server. Free holds up to 3 agent connections, with 2 running at once; the numbers grow by plan (step 5).
  • The agent takes no paid seat. Its member account is not billed and does not count toward your plan’s member limit.
  • Reads are never capped. Agent write actions have monthly caps on Free; Professional and up remove them. See AI agent action limits.
  • The assistant thinks on its own account. Your Claude or ChatGPT subscription pays for the assistant’s reasoning. Onplana’s AI tokens are spent when Onplana’s own AI does the writing: the in-app assistant, the Onplana Agent, and agent tools such as a status digest, a project summary or release notes. See Understand your AI token budget.

2. Make the agent a member (about 15 minutes)

Section titled “2. Make the agent a member (about 15 minutes)”

The first time an assistant connects, Onplana adds it to your workspace as an agent persona: a member with an agent badge, so everything it does is shown under its own name, never silently under yours.

  1. Give it the projects it should work in. The persona starts with the Member role, so it sees only the projects it belongs to. Add it from the project’s … menu, Manage project, then Invite members.

  2. Know whose permissions it carries. An agent acts with the lower of two roles: its persona’s and the person driving it. Raising the persona to Admin never gives a Member teammate Admin rights through their assistant.

  3. Decide who may use agents (the allow-list). In Organization Settings → AI and Storage, under External AI agents:

    • Allow external AI agents: off means only the Owner can use agents.
    • Who may use them: Everyone can use agents, Only selected users / teams, or Everyone except selected users / teams. A person you remove is stopped on their agent’s next call.
    • Max registered connections and Per-user cap, if you want lower limits than your plan’s.

  4. Let several agents share the work. When two agents, or two sessions of one agent, work the same project, each one leases the task it starts, so the other is handed a different task. There is nothing to set up, and a lease never locks a task for people. See Run multiple agents on one project.

Want the agent to react to a comment in seconds instead of on its next check-in? Run the relay on the machine where the agent runs.

3. Assign work and approve what comes back (about an hour)

Section titled “3. Assign work and approve what comes back (about an hour)”

  1. Assign a task to the agent the way you assign a person, and write the brief in the description: what done looks like, the constraints, and where the result goes. See Assign work to an AI agent.

  2. Let it work. The agent picks the task up, sets it to In progress, and posts progress in the task’s Agent thread, where you can answer or redirect it. See Talk to an agent in a task thread.

  3. Approve the draft. When the Onplana Agent (started with Run with Onplana Agent on a task) produces a document, report or list, the draft waits in the Agent review inbox. Edit it, approve it, or reject it with a note the agent reads next time. See Review and approve agent output.

The governance that stops a silent write

Section titled “The governance that stops a silent write”

Be clear about the default: a connected agent’s changes apply straight away, within its permissions, and each one is recorded under the agent’s name. To put a person in front of a change, use one of these:

  • Hold a call for approval. In Settings → Agents → Agent Policies, a rule can allow a call, deny it, or hold it until a signed-in person approves it. API tokens and agent connections cannot approve, so an agent cannot approve its own call. The PMO governance and Consulting starter packs add approval rules for finance and client projects.
  • Ask for a proposal instead of a change. The agent’s propose tools (status change, deadline extension, assignment, dependency, task split, milestone completion) leave a suggestion on the Suggestions tab of the Agent review inbox. Nothing changes until a person accepts it. Put “propose, do not apply” in the brief.
  • Decide what agents may delete. Deleting a task, project, list or document is on until you save your own choice, and each one goes to the recycle bin. See Control what agents can delete.
  • Stop everything at once. Turning off Allow external AI agents stops every agent except the Owner’s.

4. Budget the AI and the agent-days (about 15 minutes)

Section titled “4. Budget the AI and the agent-days (about 15 minutes)”

An agent setup draws on two prepaid balances, both on Organization Settings → Billing. Neither bills automatically: when one runs out, that thing stops with a clear message until you top it up.

AI tokens pay for Onplana’s own AI (step 1 lists what uses them). Paid plans include an allowance per billable seat every month; Free includes a one-time bonus. Agents are not billable seats, so connecting one does not change the allowance. You can buy AI credit on any plan, and an Owner can cap each person’s share of the pool with the Per-user AI cap. See Understand your AI token budget.

Agent-days pay only for hosted execution, where Onplana starts the agent for you in your AI provider’s sandbox on your own provider key. An agent-day is one connection for one UTC day, however many runs it makes. An agent you run yourself, including through the relay, uses none. Included days renew monthly and do not roll over; bought days cost USD 1.00 each and never expire. A daily schedule needs about 30 a month. See Run a hosted agent on your own API key.

Example. A 10-seat Business workspace gets 15,000,000 AI tokens a month (10 seats at 1,500,000) and 15 hosted agent-days a month, before any purchase.

5. Which steps need Business

Section titled “5. Which steps need Business”

None of steps 1 to 4 needs a paid plan. What changes by plan is how much agent work the workspace can carry, and which areas agents can reach:

For your agent setup Free Professional Business
Connect Claude, ChatGPT, Copilot, Gemini Enterprise or any MCP client Yes Yes Yes
Allow-list, Agent Policies, approvals, review inbox, leasing Yes Yes Yes
Agent connections that may exist 3 10 25
Agents running at once 2 3 5
Agent write actions Capped each month No monthly cap No monthly cap
AI tokens included 100,000 one time 1,000,000 per billable seat a month 1,500,000 per billable seat a month
Hosted agent-days included a month 0 5 15
AI risk detection, portfolio insights, schedule suggestions No Yes Yes
Portfolios, goals and OKRs, webhooks, Monte Carlo schedule risk (for people and their agents) No No Yes
Price per seat a month USD 0 USD 12 USD 20

Choose Business when more than 3 agents work at the same time, when the team keeps more than 10 agent connections, when you want more hosted agent-days included than Professional’s 5, or when agents should work across portfolios and goals, run Monte Carlo schedule risk, or push events to your own systems through webhooks.

Professional is enough when one or two assistants work project tasks and you want no monthly cap on their writes.

What Business does not include: the Agent Activity log of every agent operation, audit logs, single sign-on and SCIM, and the proposal governance pipeline come with Enterprise. See Review what AI agents did and Compare plans and upgrade.

FAQs

Section titled “FAQs”

Do we need Business before we start? No. Connect, assign and approve on the plan you have. The table in step 5 shows when the setup outgrows it.

Does each agent cost a seat? No. An agent persona is not a billable seat and does not count toward the member limit. On Business, what limits agents is 25 connections, with 5 running at once.

Can the agent change things without anyone knowing? No. Every change is recorded under the agent’s name. If a change should wait for a person, use an approval rule or ask for proposals (step 3).

Section titled “Related”