Do coding agents recommend Tavily?
Tavily was chosen in 12% of 639 judged AI search sessions, ranking fourth. Measured with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code.
Tavily was chosen in 12% of 639 judged AI search sessions, ranking fourth. It was also raised as a candidate in 253 further sessions without being chosen.
This page reports what happened when Claude Code, Codex, Cursor, Grok Build CLI and Muse Code had to solve a problem in AI search inside a realistic codebase. Not what a chat assistant says about Tavily. What an agent actually installed.
The numbers
| Category | AI search |
| Sessions in the category | 639 |
| Sessions where Tavily was chosen | 79 |
| Install share | 12% |
| Rank in category | 4 of 42 |
| Codebases it won in | 9 |
| Raised as a candidate, not chosen | 253 |
| Chosen when considered | 24% |
| Site | tavily.com |
By agent
The agents treat Tavily very differently. Muse Code chose it in 32% of its sessions in this category and Claude Code in 1%, a spread of 31 points.
| Agent | Sessions | Chose Tavily | Share |
|---|---|---|---|
| Claude Code | 140 | 1 | 1% |
| Codex | 192 | 15 | 8% |
| Cursor | 136 | 17 | 13% |
| Grok Build CLI | 35 | 3 | 9% |
| Muse Code | 136 | 43 | 32% |
A single blended install share would report the midpoint and hide both ends. If most of your users run Claude Code, your real position here is 1%, not 12%.
By who was asking
Tavily does much better with one kind of buyer than another. It won 22% of sessions asked as junior developer and 6% of those asked as senior engineer.
| Who is asking | Sessions | Chose Tavily | Share |
|---|---|---|---|
| Vibe coder | 106 | 8 | 8% |
| Junior developer | 208 | 46 | 22% |
| Senior engineer | 157 | 10 | 6% |
| Enterprise team | 168 | 15 | 9% |
What Tavily was up against
The full ranking in AI search, from the same sessions:
| # | Product | Runs won | Share |
|---|---|---|---|
| 1 | Exa | 92 | 14% |
| 2 | Brave Search API | 89 | 14% |
| 3 | Anthropic web search | 86 | 13% |
| 4 | Tavily (this page) | 79 | 12% |
| 5 | OpenAI web search | 32 | 5% |
| 6 | Built in-house (no product adopted) | 32 | 5% |
| 7 | Firecrawl | 31 | 5% |
| 8 | Parallel | 29 | 5% |
What this means
This is an integration problem, not a presence problem. Agents raised Tavily in 253 sessions and chose it in 79, so it reaches the shortlist and then loses. Something at the last step is costing the session, and in our data that is usually a quickstart that does not run when pasted, documentation describing an interface that changed, or a package name that does not match the product name.
That is the cheaper of the two problems to have. The reason is written down in each losing transcript.
Where these numbers come from
The 639 sessions in AI search are part of a published set of 15,051, run with real coding agents inside realistic codebases and judged blind. The full method is on one page: how we measured this.
Every AI search run can be replayed on the board.
If you work on Tavily: the judge recorded a reason for every session where it was raised and passed over. Those reasons are in the transcripts.
<!-- generated by scripts/write-data-pages.mjs -->
Common questions
Do coding agents recommend Tavily?
Yes. Tavily was chosen in 79 of the 639 judged sessions in AI search, a 12% install share, ranking fourth in its category.
Does Claude Code recommend Tavily?
In 1 of the 140 sessions in AI search run with Claude Code, which is 1%.
Do different coding agents treat Tavily differently?
Yes, and by a wide margin. Muse Code chose it in 32% of its runs and Claude Code in 1%.
How was this measured?
Real coding agents at pinned versions were run in sandboxes inside 92 realistic codebases and asked to solve real tasks. A simulated project owner approved or questioned each recommendation before any code was written, and a judge from a model family that builds none of the agents read every session blind.
How often is Tavily considered but not chosen?
It was raised as a candidate in 253 sessions without being chosen, and chosen in 79. That is a 24% conversion from considered to chosen.
Where this comes from
Armature ran 15,051 judged sessions with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code inside 92 realistic codebases, and published every run. The numbers on this page come from that work.