Repository navigation
Yila: a self-hosted research agent built on eve, live in production #5212
liuxiaopai-ai
started this conversation in
Show and tell
Replies: 1 comment
|
Congrats to the entire Vercel & Eve team on eve 1.0 GA! 🎉 Quick production milestone update since we shared this thread in October: Yila AI has now grown to $50K MRR (~$600K ARR), 100% bootstrapped, running multi-tenant academic research workloads on A few things that have mattered most taking
Live interactive examples (no login needed): yila.ai/use-cases If the Eve / Vercel team is collecting real-world eve 1.0 production case studies, we'd be glad to share our architecture lessons and numbers in more detail — feel free to reach out at |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi everyone! I’m Xiaopai, the founder of Yila. We’ve built a research-agent product on eve, and it’s live in production.
Yila helps researchers move from a question, a set of papers, or a dataset to work they can inspect and keep: a literature review with citations, an evidence matrix, an analysis script, scientific figures, or an editable presentation.
Our setup: we use the open-source eve framework and self-host the application. We don’t use Vercel hosting or AI Gateway. I wanted to share what that looks like as an actual research product.
Three examples you can explore
The screenshots below come from our public case collection, where you can follow the tasks step by step. The numbers describe those individual examples.
1. From a research question to a literature review
In a review about retrieval-augmented generation and factuality, Yila gathers 60 candidate papers, screens the evidence, and organizes a matrix of study designs, measures, findings, and limitations. The final synthesis includes 31 sources, with 25 verified DOIs, and keeps conflicting findings and gaps visible.
The useful part is being able to move between the search, the evidence, and the writing. The deliverables include a Word draft, Markdown source, an Excel evidence matrix, and a CSV of search records.
Scholarly search in the research workbench. Explore the literature-review case →
2. From a data table to figures and reproducible code
A transcriptomics example starts with a processed public dataset from NCBI GEO (GSE52778). Yila inspects the columns and comparison direction, then produces a volcano plot, PCA visualization, result tables, and a Python script to reproduce the analysis.
In this example, the chosen thresholds identify 316 differentially expressed genes. The write-up separates direct observations from biological hypotheses and explains the limitations of starting from processed data.
The generated volcano plot; the case also includes an editable SVG and reproduction script. Explore the data-analysis case →
3. From a teaching topic to an editable slide deck
For a lecture on prompts and skills, Yila proposes teaching objectives and an outline, waits for confirmation, then builds a 12-slide deck using comparison tables, process diagrams, and case examples. It renders and checks the slides and exports an editable PPTX.
The screenshot shows the Chinese-language example alongside the workbench’s slide preview. Yila also supports paper-to-slides tasks, scientific figures, and academic posters.
An editable lecture deck in the artifact preview. Explore the presentation case →
Where eve fits
Research work often spans many tool calls and follow-up messages. The agent needs to keep working through files, ask for input when necessary, and carry the result into the next turn.
We use eve for the agent execution layer: the model/tool loop, durable sessions, streaming, and human-in-the-loop interaction. Around it, Yila adds the research workbench and domain-specific capabilities:
We keep product records and workflow execution state separate. That distinction matters when a browser reconnects or a long-running task waits for a researcher’s input.
What we’ve learned building it
Thanks to the eve team and contributors for building the foundation. I’d love to hear from others taking eve into production, especially for long-running, file-heavy workflows.
Try Yila · Browse the public examples
Happy to discuss the integration here. Which would be most useful for a follow-up technical write-up: self-hosting, task continuity, or research artifact delivery?
All reactions