# Two teammates who never forget a check-in. > Two AI teammates ship with every OKRGoal workspace. The Coach guards quality and the Champion keeps the rhythm. They propose, you approve, every message cites a rule, and they still work with AI switched off. This is the Markdown version of https://okrgoal.com/product/agents *The agents* Every OKRGoal workspace comes with two AI members. The **OKR Coach** guards the quality of your OKRs. The **OKR Champion** guards the rhythm that keeps them alive. They start the conversation, in the browser and in your chat apps, and every message tells you which rule it comes from. [Start free](https://app.okrgoal.com/sign-up) · [Try the live demo](https://demo.okrgoal.com/) ### OKR Coach Guards quality Watching 42 goals in 6 spaces. ### OKR Champion Guards the rhythm Watching 42 goals in 6 spaces. *Members, not features* ## They have names, a scope and a job. The Coach and the Champion are members of your workspace. They appear in the activity feed. Their actions are in the audit log under their own names. And they never count as seats, on any plan. ### Seeded in every workspace Both are there the moment the workspace exists. Nothing to install. ### Scoped to what you give them Each sees only the spaces, goals and KPI trees it is bound to. Never the whole workspace. ### Never a seat Agents and guests are free on every plan. *Guards quality* ## The OKR Coach reads every draft so the quarter does not start on a bad one. Its whole vocabulary is the method. It reviews every objective and key result as it is written, guards the publish gates, and speaks at the moments a good facilitator would. When it acts - **Continuously** On every change to a goal, key result or alignment link - **At each phase** It explains what is blocking the phase, and why - **Nightly** With an AI provider connected, a sweep for duplicates, conflicts and drift across the whole tree - **When asked** An administrator can run it at any time, and read every run's log *Guards the rhythm* ## The OKR Champion does the chasing, so nobody has to be the nag. It reminds the goal's champion before a check-in is due, then escalates when it is late, always where the person can see it. It holds every blocker to the next check-in. It announces the weekly session, assembles the digest for the coordinator to post, keeps the streak, watches every KPI, and tells the facilitator when the quarterly review is due. Its week - **Every hour** Works out who is due what, on which channel - **Every morning** A summary in each person's own time zone. Checks staleness, blockers and KPIs - **Every week** Announces the session, assembles the digest, updates the streak - **Every cycle** Counts down to the publication date, then tells the facilitator the review is due *The escalation ladder* ## Escalated, never ambushed. A missed check-in climbs a ladder, one step at a time, and the person being escalated past can see every step. Nobody is reported behind their back. - **On the day:** **The goal's champion** - **1 day late:** **The champion again, on their main channel** - **After 3 days:** **The reviewer, if the goal has one.** The goal now shows **outdated** - **7 days late:** **The space coordinator** - **14 days late:** **The sponsor, only if your workspace adds them.** Otherwise it stops at the coordinator Every step and every delay is a setting. A late check-in can first be offered to a teammate as help, before it goes anywhere more senior: optional, and off by default. Holidays pause the clock without breaking the streak, and a member on leave names a delegate. *Every message cites a rule* ## Argue with the rule, not the robot. Every proactive message is a recorded nudge with four things attached: the rule that sent it, the channel it went to, where it sits on the escalation ladder, and the reason if it was held back. Click the rule and you are reading it in the method. *Safe by design* ## Built for the people who say no to AI. ### They propose. You approve. By default every change an agent wants to make becomes a proposal in your To do list, under Updates. Apply it or dismiss it, one at a time or in bulk. ### Sandbox mode Run an agent in a full dry run. It shows you everything it would do and commits nothing. ### Least privilege An agent gets explicit access to named spaces, goals and KPI trees. There is no service account with access to everything. ### Hard cost caps Every step is metered. A cap halts a run mid-flight, with a clear log line. An agent cannot run up an unbounded bill. ### Quiet by design One nudge per subject per day, unless it escalates. At most five per person per week, and the rest wait for the morning summary. An escalation to someone else is never held back. Champion two goals? You hear once, and check in on them one after another. ### Fully audited Every action is in the audit log under the agent's own name, next to who approved it. Trust one agent enough? Give it direct writes, scoped to the spaces you name. It is your call, per agent, and it is off until you turn it on. *With AI and without it* ## Turn the AI off. They still work. The rules behind every escalation, gate, score and diagnostic, and every nudge but one, are code, not a prompt. Without an AI provider, the agents still do every part of their job that needs a decision. They lose the drafting, the language and the one nightly sweep that needs a model to judge meaning. | Part of the job | With AI off | With an AI provider | |---|---|---| | Quality checks and gates | All twenty-six checks, all six gates | Plus a rewrite of each failing line | | Reminders and escalations | Yes | Yes | | Weekly digest | From a template | Written for this week | | Overdue check-in | A blank check-in, ready to fill | A status, a confidence and a line of story, drafted from where each key result stands | | Recovery OKR | Drafted from the drivers that are short | Plus a better-worded objective | | Alignment review | Structural gaps | Plus duplicates, better parents and hidden dependencies | | Closing diagnostic | The verdict and its prescription | Plus specifics from this cycle | Every change we make is tested with the AI switched off, so this table stays true. ## Questions ### Which AI does it use? Yours. Anthropic, OpenAI, Gemini, OpenRouter, Ollama, or any OpenAI-compatible endpoint. Keys can belong to the instance, the workspace, or a person once an admin allows personal keys. Point it at a local model and nothing leaves your network. ### Can I create more agents? Yes. A workspace can create its own agents, each with a persona, instructions, a schedule, a scope and a write policy. The Coach and the Champion are the two that ship with every workspace. A custom agent needs an AI provider, because it has no rules of its own to fall back on. ### Will people get spammed? No. Nudges are a budget. One per subject per day, at most five per person per week, quiet hours in each person's own time zone, and a snooze for every message, in the browser or with a typed command in chat. A snooze quietens the message. It never hides the obligation. ### What stops an agent seeing something it should not? The same thing that stops a person. Every agent action goes through the same permission check and the same database isolation as a click in the browser, using the agent's own narrow access. ### What data reaches the AI provider? What you allow. Admins choose how much context may leave, whether email addresses and phone numbers are redacted, and which hosts may be reached. A local model means nothing leaves at all. ## Let them run your next check-in. [Start free](https://app.okrgoal.com/sign-up) · [Try the live demo](https://demo.okrgoal.com/) --- OKRGoal, from OKR Institute. A map of this site for AI agents: https://okrgoal.com/llms.txt