ellmer
ellmer stopped being a chat wrapper and started shipping the parts production LLM code needs
A side-by-side editorial comparison of btw and Firecrawl — release velocity, themes, recent moves, and the top alternatives to consider.
btw is turning into an agentic R harness that no longer needs you to be in R
btw assembles context about an R session — packages, documentation, files, data frames — and hands it to an LLM through ellmer, with btw_app() as a chat interface. Over the last year it has grown well past context assembly: LLMs can document, check, test and measure coverage of a package, read CLAUDE.md and AGENTS.md as project context, fetch skills from packages or GitHub, and inspect the source of any installed namespace. Much of this is now reachable from a terminal CLI rather than only from an R prompt.
Firecrawl stopped selling pages and started selling answers — now it is giving the corpus away.
Firecrawl has spent four months converting a scraping API into an answer-retrieval layer for agents. Question, Highlights and the excerpt-scoring rebuild of /search all trade full-page delivery for the specific lines that answer a query, each pitched on token cost rather than coverage. Alongside that it has started owning corpora outright — the Research Index now spans 3M+ arXiv papers and 41M+ life-sciences papers — and /monitor turns crawling into a subscribable event stream.
btw assembles context about an R session — packages, documentation, files, data frames — and hands it to an LLM through ellmer, with btw_app() as a chat interface. Over the last year it has grown well past context assembly: LLMs can document, check, test and measure coverage of a package, read CLAUDE.md and AGENTS.md as project context, fetch skills from packages or GitHub, and inspect the source of any installed namespace. Much of this is now reachable from a terminal CLI rather than only from an R prompt.
The direction is from describing a session to operating on it, and from inside R to outside it. Each release adds either a tool group that lets a model do something (document, check, test, cover; read namespace source; fetch skill resources) or a CLI command that removes the need to start R first. The 1.2.0 tool renaming — session becoming sessioninfo, search becoming cran, files_read_text_file becoming files_read — reads as the naming cleanup you do when you expect a lot more tools to follow.
The CLI has been absorbing one tool family per release (skills, then pkg desc and pkg src) while the R-side tool groups stay ahead of it, so the next releases likely continue exposing existing tool groups as terminal commands rather than adding new capabilities.
Firecrawl has spent four months converting a scraping API into an answer-retrieval layer for agents. Question, Highlights and the excerpt-scoring rebuild of /search all trade full-page delivery for the specific lines that answer a query, each pitched on token cost rather than coverage. Alongside that it has started owning corpora outright — the Research Index now spans 3M+ arXiv papers and 41M+ life-sciences papers — and /monitor turns crawling into a subscribable event stream.
The centre of gravity is moving from generic crawl infrastructure to curated indexes with published benchmark numbers attached, and now to giving those indexes away. Every recent release argues the same point in a different register: the crawler should return the smallest correct thing, and Firecrawl should already have it indexed. Free access to Research Index converts a metered data product into a distribution channel for the paid scraping and monitoring endpoints around it.
Expect a third vertical index after AI/ML and life sciences — the pattern of a benchmark claim, daily refresh and API-plus-MCP-plus-CLI availability is now a repeatable template. Whether the free tier stays free once query volume lands is the open question the entries do not answer.
Other ai-assistants products tracked by Sparkpulse, ranked by recent ship velocity. Each card links to a full editorial trajectory and lets you pivot into a head-to-head comparison with either btw or Firecrawl.
ellmer stopped being a chat wrapper and started shipping the parts production LLM code needs
tfevents logs TensorBoard events from R, and this release only changes who maintains it.
safetensors for R changes hands with no code change to show for it.
torchdatasets ships custodial work as mlverse gathers its torch satellites under one maintainer.
A curated catalogue of published hyperparameter search spaces, now reaching deep neural networks
Hyperband tuning for mlr3, now built on an asynchronous backend it treats as mandatory
See all btw alternatives → · See all Firecrawl alternatives →
Latest ship moves from both products, interleaved chronologically. ⚡ = editorial spark.
They serve adjacent needs but don't currently overlap on shipped themes. Firecrawl is currently shipping more aggressively (velocity 7.5 vs 2.5), with 2 editorial sparks in the last 30 days against 0. See the at-a-glance table above for a side-by-side breakdown of velocity, recent sparks, and editorial themes.
Sparkpulse doesn't pick a winner — we score release velocity, not feature parity. Firecrawl is currently shipping more aggressively (velocity 7.5 vs 2.5), with 2 editorial sparks in the last 30 days against 0. For your specific use case, the alternatives sections above list other ai-assistants products to evaluate alongside.
Top btw alternatives in ai-assistants are ranked by recent ship velocity. Browse the "btw alternatives" section above for the current picks, or visit /alternatives/btw-r for the full list with editorial commentary on each.
Top Firecrawl alternatives in ai-assistants are ranked by recent ship velocity. Browse the "Firecrawl alternatives" section above for the current picks, or visit /alternatives/firecrawl for the full list with editorial commentary on each.