A collaborative AI scientist on your desk.
Literature · research design · experiments · LaTeX — one local workspace.
Scope note. The research loop is usable today. We do not claim an unattended “auto-publish science” product. PrismNext is built for serious co-driving: the agent can advance the work; you keep the gates.
PrismNext is a local-first desktop workspace for academic writing and research practice. It is not a generic coding agent with a chat pane bolted on, and it is not a black-box pipeline that promises overnight manuscripts.
It keeps the loop read → design → run → write → review in one application, with research objects (library, brief, plans, experiments, TeX sources) that live on your disk.
- Researchers (PhD, MSc, and beyond) who live in papers, experiments, and citations
- Small writing teams sharing a repository, with Git / worktrees and controlled AI assistance
- Authors who write LaTeX in earnest — SyncTeX, compile, diffs — not Markdown-only substitutes
- Privacy-minded users — manuscripts and library stay local; you supply your own model API keys
If you only need casual literature Q&A, this may be more structure than you want.
If you want a tool open all day while you work, that is the intended use.
| Generic coding agents | Lit-chat / “auto AI scientist” | PrismNext | |
|---|---|---|---|
| Focus | Files + chat + tools | Retrieval or overnight pipelines | Research objects with lifecycle |
| Control | Little scholarly structure | Weak manuscript / integrity gates | Plan consent, approve-to-build, permission modes |
| Data | Cloud / IDE-centric | Often SaaS-first | Local-first + your API keys |
| Outcome | “Edit the .tex” |
“Find papers / generate claims” | Read → design → run → write → cite → compile |
One principle: the agent advances; you keep the veto.
- Project library (SQLite) and PDF reading — not disposable chat attachments
- Metadata enrichment (Crossref, arXiv, OpenAlex, …)
- BibTeX import / export, citation staging, citation health
- Search, then stage into the local library so the shelf grows with the project
- Research Brief — living design notes that travel with the project
- Plan workflow — Build | Plan; consent before the agent enters Plan (
⌥P/Alt+Pto toggle) - Draft → Approve & Build → checklist — fewer uncontrolled “chat wandered off” edits
- Experiment workspaces, run logs, artifact snapshots
- Provenance — which command produced a given plot or result (useful for Methods)
- TeX workspace (outline, find, language support)
- pdf.js preview + SyncTeX (both directions)
- Tectonic (default) or TeXLive;
% !TEX program/% !TEX root - Proposed Changes — merge view; accept / reject per change or in bulk
- Export PDF or source zip
- Multi-tab streaming chat
- Orchestrator + experts (literature, design, methods, structure, peer review, …)
- Skills · slash commands · knowledge modules · project rules
- Permission modes: Ask / Edit-auto / Auto / Readonly
| Mode | Role |
|---|---|
| Human-led | You write and compile; ask when stuck |
| Co-drive | Agent advances design / runs / prose; you approve major beats |
| AI-led | Maximize the work surface; intervene lightly via the capsule bar |
- Git: status, stage, diff, commit, branches, merge, stash
- Worktrees for parallel writing contexts
- Terminal + AI bash
- In-app browser
- Templates: paper / thesis / beamer / poster / CV / letter
- UI languages: English · 简体中文 · 繁體中文(香港)
- Packaged builds support in-app updates
Download macOS, Windows, or Linux (AppImage) builds from GitHub Releases (or the download page when hosted).
On macOS, if Gatekeeper reports the app as “damaged,” clear quarantine once, then reopen:
xattr -cr /path/to/PrismNext.app
Choose a folder as the project root. PrismNext maintains .prismnext/ there (library, brief, experiments, compile cache, …) — on your machine.
Settings (⌘, / Ctrl+,) → provider and API key (bring your own key).
There is no step that uploads your thesis to a Prism cloud.
- Add papers to the library (or search → stage)
- Capture problem and path in Brief / Plan
- Write in TeX with live PDF preview
- Attach experiment runs and artifacts when needed
- Chat when stuck; use Plan and permission modes for large moves
- Review Proposed Changes — keep only what you intend
- Manuscript, library, experiments, agent configuration → your machine
- Model calls → your keys
- Optional network use (enrichment, search MCP, …) → explicit, not a silent project sync
Designed for unpublished data, sensitive drafts, and authors who do not want an entire project living only in a SaaS.
Brand and loop diagrams live under assets/. Product shots used by the download site are under website/assets/.
Pull requests with additional UI captures (welcome, TeX + PDF, library, Plan consent) are welcome.
We are deepening co-drive:
- Bounded, auditable topic discovery into the local library
- Clearer evidence / stance snapshots (not “consensus declared”)
- Sharper Human / AI / Co drive modes
Not current goals: default overnight unattended discovery, auto-submission, or replacing the scientist.
Issues and PRs that improve the research loop, consent gates, or local-first engineering are welcome. See CONTRIBUTING.md, SUPPORT.md, and CODE_OF_CONDUCT.md. Security reports: SECURITY.md.
Licensed under the Apache License 2.0. Copyright © 2026 yibocat — see NOTICE.
PrismNext — co-drive serious research, locally.