How Agent Rank works
We score every agent in our catalog on six dimensions, sum to a 0–100 aggregate, and classify into tiers. The formula is public, deterministic, and re-evaluated on every deploy.
The six dimensions
Autonomy fit
How we compute it: Mapped from the agent's autonomy level: Assistant → 7, Semi-autonomous → 8, Autonomous → 9.
Why: Higher autonomy isn't always better, but it's what buyers ask about. The +2-point spread captures that distinction without overweighting it.
Capabilities
How we compute it: Two points per native capability (tool use, code execution, RAG, memory, vision, voice, browser use, multi-agent). Five or more capabilities caps the dimension at 10.
Why: Breadth across capability types is a stronger predictor of fit than depth within one type.
Integrations
How we compute it: Two points per distinct integration category (IDE, CRM, ticketing, messaging, storage, etc.). Five or more categories caps at 10.
Why: An agent with 10 sales-only integrations is less broadly useful than one with five integrations across three categories. We reward category breadth, not raw count.
Pricing value
How we compute it: Open source: 10. Free: 9. Under $25/mo: 8. Under $50/mo: 7. Under $100/mo: 6. Under $200/mo: 5. Higher or contact-sales: 3–4.
Why: Pricing value isn't 'cheapest wins.' It's 'do I get fair power for what I pay.' Open source scores highest because licensing is the unlimited tier.
Polish & maturity
How we compute it: Baseline of 4, plus 1 each for: a published version number, featured status, ≥50K views, ≥150K views, having a product preview image or video, having multiple pricing tiers. Capped at 10.
Why: Mature, well-known agents with thoughtful product surface score higher. These are the signals that a real engineering team is behind the product.
Verifiability
How we compute it: Baseline of 3, plus 4 for our Verified badge, 2 for authored pricing tiers, 1 for authored pros/cons. Capped at 10.
Why: Verifiability captures how much editorial work we've done on the listing. An agent we've personally vetted, with structured pricing and pros/cons, beats a placeholder record.
Aggregate score
The six per-dimension scores (each 1–10) are summed, divided by 60, multiplied by 100, and rounded. The result is the agent's Agent Rank score.
Tier classification:
- S-tier: 85 or higher — top of the category, well-rounded across all six dimensions
- A-tier: 72–84 — strong, with at most one weak dimension
- B-tier: 58–71 — solid, with multiple trade-offs to consider
- C-tier: below 58 — niche or early-stage; check the breakdown carefully
What the score is not
Agent Rank measures fit across our six dimensions — it doesn't measure your fit. A C-tier agent can be perfect for a specific workflow that values one dimension we underweight. The score is a starting point; the breakdown table on each agent page is where the real decision lives.
Editorial overrides
For agents where the automated score visibly diverges from hands-on experience, the editorial team can override a single dimension. All overrides are public — they appear in the per-dimension table on the agent page with a small note.
Refresh cadence
The score is recomputed on every deploy of this site. When we add a pricing tier, ship a new pros/cons block, or update a capability list, the score moves the next time we deploy. Major catalog updates ship weekly; structural changes to the formula are versioned and announced in our newsletter.
User reviews are separate
Agent Rank is the editorial score. Community user reviews surface separately on each agent page with their own aggregate. We don't blend them — they answer different questions ("is this agent well-built" vs. "did real users like it").