Diagnostic
Paste your OKR. Get a structured critique in seconds: what's broken, why, and three rewrites you can use immediately. Your API key never leaves your browser.
Runs locally before the LLM call. Output-verb detection, timebox regex, placeholder check. Instant pre-score, no key required.
Try an example
Paste OKR
Key stored in browser only. One OKR set at a time.
~$0.002 per analysis with GPT-4o-mini. ~$0.01 with Claude Sonnet. Cmd+Enter to submit.
~$0.02-0.04 per session. Conversation stays in your browser only.
Rule engine flags
Findings
Getting LLM analysis
Suggested rewrites
Set-level analysis
The "So What?" test
OKR Orca scores against a seven-criterion rubric. Three criteria apply to the Objective. Two criteria apply to each Key Result. Two apply to the set as a whole.
The core principle: "Who does what by how much." An outcome is a measurable change in behaviour. Outputs are not outcomes. Impact (revenue, profit) is too far removed. Key Results live at the outcome layer.
For every KR, ask: (1) If all KRs are green, is the Objective obviously achieved? (2) If this KR turns red, does it signal a real problem? (3) Does the team actually control this metric? Any "no" means the KR needs rewriting.
Common anti-patterns: Output-as-KR. Impact-as-KR (revenue targets). Vanity metrics (engagement without "who + by how much"). Placeholders. Binary milestones (100% migrated). Task lists disguised as KRs.
No. Your OKR text is sent directly from your browser to OpenAI or Anthropic. OKR Orca has no backend. The source is public and you can verify this yourself.
In your browser's localStorage only. It is never transmitted to okrorca.com or any third party. You can clear it at any time via the "change" link in the nav.
OpenAI: GPT-4o-mini (default, cheapest), GPT-4o. Anthropic: Claude Sonnet 4.6 (default), Claude Haiku 3.5. The rule-engine pre-score runs locally and is free regardless of key.
With GPT-4o-mini roughly $0.001 to $0.003 per analysis. With Claude Sonnet 4.6 roughly $0.008 to $0.015. The estimate shown on the input panel is indicative only.
Yes. The rule-engine pre-score runs instantly in the browser with no key required. You get output-verb flags, timebox checks, placeholder detection, and a pre-score. LLM rewrite suggestions require a key.
A heuristic score (0-100) computed locally in under 100ms. It checks for output verbs in KRs, missing timeboxes in the Objective, placeholder strings, and the presence of baseline and target numbers. It is a fast signal, not a replacement for the full LLM analysis.
Coach mode is the second tab in the input panel. Instead of pasting an existing OKR, you start a conversation. The tool asks one focused question at a time, surfaces the gap between what you want to achieve and what you have written, and helps you arrive at an OKR in your own words. When the draft is ready, you can send it straight to Diagnose for a rubric score.
I'm Frederik Metz, an agile coach in Munich. I've reviewed several hundred OKRs over the last few years, mostly from product teams trying to get the rubric right and getting tangled by output-disguised-as-outcome KRs.
OKR Orca is the diagnostic I wish my teams had earlier. The 7-criterion rubric is the one I use in coaching conversations. It's opinionated, but it's the opinion that consistently produces OKRs that move outcomes, not OKRs that produce planning theatre.
The tool is free. Your key stays in your browser. No signup, no tracking. If you find a bug or want a feature, the source is on GitHub.