How we rank
Every score is out of 100, built from seven criteria with fixed weights. The weights and the evidence rules are published here so you can disagree with them.
The honest limit: at launch, every ranking is based on public information (vendor documentation, pricing pages and independent write-ups). We have not hands-on tested these tools, so every score is labelled Docs-only or Signals-only. A Tested label appears only once we have run the tool on a standard task set and kept the results.
Two rubrics
Tools that you operate (an image generator, a notes app) use the general rubric. Agents that act for you (a support agent, a browser agent) use the agent rubric, which swaps in autonomy, safety and cost predictability.
General tools (100 points)
| Criterion | Points | What we look at |
|---|---|---|
| Output quality for the job | 30 | How well the tool does the one job, judged from documented capabilities and the best third-party tests we could find. Untested tools are capped at 24 of 30. |
| Ease for a non-developer | 15 | Can a busy person get a good result in the first hour without technical help? |
| Value for money | 15 | Useful free tier, clear pricing, and no nasty renewal jumps. |
| Reliability and vendor stability | 10 | Is the vendor established, shipping, and likely to be here next year? Shutdowns cost points. |
| Privacy and data control | 10 | Clear terms, data controls, and where your content goes. |
| Integrations and ecosystem | 10 | Works with the tools people already use. |
| Momentum | 10 | Pace of real improvements in the last 90 days. |
Agents (100 points)
| Criterion | Points | What we look at |
|---|---|---|
| Task completion | 30 | Does it finish the job end to end, judged from documented capabilities? Untested agents are capped at 24 of 30. |
| Useful autonomy | 15 | Does it work without constant supervision, and does it know when to stop and ask? |
| Safety and approvals | 15 | Human approval before sending, paying or deleting; permission controls; audit trail. |
| Cost predictability | 10 | Can you predict the bill? Credit or per-outcome billing scores lower. |
| Setup effort | 10 | How much configuration before it is useful to a non-developer? |
| Maturity and availability | 10 | Generally available, in your country, and out of beta? Waitlists and previews score low. |
| Integrations and reach | 10 | Connects to the apps where your work happens. |
Evidence levels
| Level | What it means | Rule |
|---|---|---|
| Tested | We used the tool on a standard task set and kept the results. No tool has this label yet. | No cap. |
| Docs-only | Scored from the vendor's own documentation and independent third-party write-ups. We did not run it. | Quality or task score capped at 24 out of 30. |
| Signals-only | Based only on secondary sources, such as an aggregator listing or a social post. | Total capped at 70 out of 100. |
The build fails if a score breaks these caps, so a Docs-only tool cannot quietly get a perfect quality score.
Rules that protect the ranking
- No pay-for-rank. Sponsors and affiliate partners cannot buy a score, a position or a pick. The data format has no field for them, and the build rejects one.
- Sponsored and affiliate content is labelled and sits outside the rankings. See Advertise.
- One source link per tool, at minimum. Each tool page lists the sources and says whether each is a vendor page, a third party or a competitor. Competitors' comparison posts are used for leads, and we say so.
- Prices are "as reported". We link where we saw each figure. Where sources disagree we say so rather than choose one.
- Re-scored weekly. Each use case shows the date it was last reviewed. A shutdown, a big price change or a new release can move a rank.
- Shutdowns are flagged. If a tool is discontinued, it is listed under "Skip or no longer available" with a source.
Where candidates come from
New tools are found by a weekly script from public sources that allow it: the Hacker News API, GitHub search by creation date and stars, Hugging Face trending, and vendor RSS feeds. The script drafts scores from simple signals and writes a draft that a human reviews before anything is published. We do not use the Product Hunt API, whose terms bar commercial use without permission.
Known limitations
- Scores are editorial estimates. Two reasonable people could rank neighbours differently.
- Third-party roundups can be wrong, out of date or written by competitors.
- Agent features change fast and some are limited by country, such as US-only rollouts.
- We might miss a tool. Tell us at hello@toolpodium.com (placeholder address).