Largest free searchable archives of human–AI conversations, compared
Checked November 17, 2026. Sizes and limits change; each row links to its source.
Short answer: as far as we know, The Chat Archive is the largest completely free searchable archive of real human–AI conversations, with 2.7M conversations you can search, read and download with no account or limits. ChatCorpus advertises more conversations (5M+) but caps free use at 10 searches a day. WildVis is free but smaller (≈1.8M). Hugging Face hosts the 3.2M-row WildChat-4.8M dataset, but its viewer's search covers only about a third of it.
| Archive | Size | Free? | Search | Sources |
|---|---|---|---|---|
| The Chat Archive Hosts only data whose license allows redistribution | 2,689,461 conversations | Yes, completely. No account, no daily limit, no paywall | Full-text keyword search over every conversation, plus a free API and bulk downloads | 10 openly licensed research datasets (WildChat-4.8M, LMArena 140K and 55K, Search Arena, HH-RLHF, OpenAssistant, HelpSteer2, PRISM, ShareLM, Gata preview) plus consenting contributors |
| WildVis Research visualizer from the WildChat authors; does not include WildChat-4.8M | ≈1.8M conversations | Yes | Keyword and filter search; results capped at 10,000 per query | WildChat-1M and LMSYS-Chat-1M |
| ChatCorpus Larger, but not free beyond the daily cap | 5M+ conversations (as advertised) | Limited. The free account gets 10 searches a day and the first 5 matches per search; unlimited access is $29/month | Full-text search behind sign-in | WildChat, LMSYS-Chat-1M, Chatbot Arena releases, Search Arena, OpenAssistant, HH-RLHF |
| Hugging Face viewer: WildChat-4.8M A dataset viewer rather than an archive; the full dataset is a free download | 3.2M rows in the dataset | Yes, with a free Hugging Face login for search | Full-text search indexes only the first 5 GB of the 15.3 GB dataset (about a third) | WildChat-4.8M only |
How we checked
- WildVis: its open-source indexer loads
allenai/WildChat-1M(837,989 conversations after removals) alongside LMSYS-Chat-1M (1,000,000). Source code. - ChatCorpus: size and sources from its sources page; limits from its pricing page.
- Hugging Face: dataset size from the Hugging Face datasets-server; search coverage from the dataset viewer documentation (only the first 5 GB of large datasets is indexed).
- The Chat Archive: the count above is live from our database. Search covers every conversation (title plus the first 4,000 characters).
What's not counted
- Downloadable datasets without a public search interface (for example the gated WildChat-4.8M-Full, LMSYS-Chat-1M on its own, or the full Gata 2.2M set).
- Paid analytics products that model or extrapolate conversation volumes rather than publish conversations.
- About 80K of our conversations (OpenAssistant) were written by human volunteers on both sides; they are labelled as such.
Spot something out of date? Tell us and we'll correct this page.