AI built for project work,
not chatbot novelty

Both. Onplana ships with dual-provider support: Claude (Anthropic) and Azure OpenAI. Onplana operates both: one serves requests as the primary, and its operators can switch to the other, for example during a provider outage. A tier abstraction (fast / balanced / powerful) maps each request to a concrete model on the serving provider. There is no per-customer provider setting: every plan gets the same Onplana-managed dual-provider engine.

No. Onplana sends prompts to Anthropic and Azure OpenAI inference endpoints under their standard enterprise terms, which do not train on customer inputs by default. We don't fine-tune our own models on your data either. Your project data stays in your org's database, AI accesses it at request time through retrieval over the tenant's own rows.

No. Onplana operates both providers itself: provider keys live in Onplana's Azure Key Vault, Onplana decides which provider serves requests, and its operators can switch to the other when one is degraded. There is no customer-facing provider control and no bring-your-own Azure OpenAI. What you do control: a monthly AI cost cap with warn and block modes, and a full audit trail of every AI action in your org.

Yes. The core AI suite (chat, plan generation, status reports, NL parsing, suggestions, Run with Onplana Agent) ships on every plan including Free. What varies is the AI token amount, sized by plan tier and seat count: paid plans get a monthly allowance per seat that resets each calendar month, and Free gets a one-time bonus that does not renew. When you need more, you buy prepaid AI credit (purchased credit expires 90 days after purchase) or upgrade. MCP agent connections are also on every plan (the concurrent-connection cap scales with tier: Free 2, Pro 3, Business 5, Enterprise 10, Enterprise+ unlimited). Separately, the write actions an agent performs per month are capped on Free and unlimited on Pro and up. Advanced AI (risk detection, portfolio insights, leveling and EVM commentary) starts at BUSINESS. The gating is real: the feature flags in the codebase actually enforce it.

Retrieval-augmented at request time. The chat panel passes a typed tool catalog to the model, the model calls tools (list_tasks, get_project, list_risks, etc.) scoped to the calling org's X-Organization-Id, and the tools return only that org's rows. No vector embeddings sit in a shared index, no cross-tenant leakage path. Every tool call is audit-logged on AiOperation with input, output, and the requesting user.

Not behind a confirm-every-write dialog, and the precise answer is stronger than a vague reassurance. Onplana does not put an approve step in front of every AI mutation by default. Three things bound it instead, all enforced server-side. AI inherits the permissions of the person it acts for, so it can never do what that person could not do by hand. Deleting a task, a project, a list, or a document is governed per org: each sends what it deletes to the recycle bin, and an admin can switch each one off, after which it is refused if called and an external agent no longer sees it in its tool list. Those four are the only delete tools an external agent connection offers. Every AI action is recorded as an AiOperation row and can be undone. If you do want a human gate on a particular operation, an agent policy rule set to approve blocks that call until a person approves or rejects it.

Paid plans include AI tokens every month, per billable seat, at no extra cost, with a larger allowance on higher tiers. The allowance resets each calendar month and unused tokens do not roll over. Free includes a one-time 100K token bonus that does not renew, and Enterprise+ is unlimited. If you need more, buy prepaid AI credit from $5 (it expires 90 days after purchase and is spent before your monthly allowance) or upgrade. Admins can set a monthly cost cap (WARN at 80%, BLOCK at 103%) to control spend, and there are no surprise overage bills, the cap rejects requests with a clear "AI quota exceeded" error rather than silently charging.

Yes, via the MCP server at mcp.onplana.com. Onplana implements Model Context Protocol with OAuth 2.1 (Dynamic Client Registration per RFC 7591) and Personal Access Token auth. Claude Code, Continue, Goose, and any other MCP-compliant client can connect and use the full Onplana tool catalog (280+ tools, same surface as the in-app AI). See /mcp-project-management for the full integration guide.

The balance is org-wide by default, so it is first come first served, which is fine for a small team and less fine for a large one. Admins can turn on a per-user share that caps each person at a set percentage of the org pool for the month. Someone who hits their share gets a distinct error telling them to ask for a raise, rather than a generic quota message, so the admin knows whether to lift one person or top up the org. Separately, admins can set a monthly cost cap in dollars that warns at 80% and blocks at 103%.

Start from the fact that you can turn it off. AI is off-switchable at the org level and, separately, each of the 51 AI surfaces has its own toggle, so adopting it is not all or nothing. Beyond that: AI acts with the permissions of the person using it and cannot exceed them, prompts go to Anthropic and Azure OpenAI under enterprise terms with no training on customer inputs, every AI action is recorded and undoable, external agent connections are admin-controlled, the four deletes they offer (a task, a project, a list, a document) go to the recycle bin and can each be switched off, and admins can cap both per-user token share and monthly spend. The honest limits are listed further down this page rather than buried.

Yes. Org owners can disable AI at the org level (no chat panel, no suggestions, no risk detection runs, no scheduled report generation). The setting is enforced server-side so even a user with a direct API call can't bypass it. SCIM-deactivated or matrix-restricted users also lose AI access through the same gating layer. It is not only a master switch: the same settings carry a separate on/off toggle for each of the 51 AI surfaces, so you can leave chat and summaries on while turning risk detection or plan generation off, and a disabled surface tells the user to ask their admin rather than failing silently.

AI is bad at: judgment calls between competing priorities, reading interpersonal team dynamics, deciding whether a sponsor is genuinely going to push back, and anything that requires sustained context across more than a few weeks of work. It's good at: surfacing the signals you didn't see, drafting a status report you would have written yourself but slower, parsing unstructured input into structure, narrating math the engine already computed. Use it as a force multiplier on the boring 60% of PM work, not a replacement for the 40% that actually needs a human PM.