nexus-chat

A local-first terminal chat app for deep research and multi-agent work. Rust + ratatui, all state on your machine — SQLite per space, files and artifacts in space directories, model-created web apps served from localhost.

Features

• Chat over any configured backend: OpenRouter, OpenAI, OpenCode Zen/Go, and Codex — one merged model list, per-model reasoning-effort control
• Deep research (/research <topic>): a conversational scoping survey → a plan of questions with why/angles/sources briefs → parallel searcher agents → synthesis → critic → verifier → writer, with a live activity view and /steer mid-run. /research! <topic> runs ungated
• Swarm (/swarm): a moderator-conducted multi-persona roundtable that iterates toward consensus
• Watches (/watch): standing research jobs that re-run daily
• Spaces: per-project context (instructions, memory, imported files, scripts, apps) with embeddings-backed semantic file search
• Skills: reusable instruction packs (SKILL.md) with sandboxed Python virtualenvs, installable from chat
• Tools for the model: nine consolidated tools — search (web/academic/ discussion), fetch_url (with PDF/YouTube extraction), batch (multi-op calls), research_lookup, files, app (build & edit web apps served on http://localhost:8642), scripts, skills, and media (image/video generation with ffmpeg transforms)
• Terminal ergonomics: markdown rendering, image display, @-file autocomplete, mouse selection → copy, context breakdown, compaction, incognito mode, per-session history

Build & run

cargo build --release
./target/release/nexus-chat

Keys come from the config file or env (OPENROUTER_API_KEY, OPENAI_API_KEY, OPENCODE_API_KEY), or /login in-app. On first launch a key is enough; models are fetched from the catalogs. Extra tooling (all optional): tesseract for local OCR, ffmpeg for video transforms, an ollama instance for local embeddings/OCR.

Commands

Type / in the composer for autocomplete. Aliases in parentheses.

Keybindings

While a research survey or plan approval is pending, Enter in that session answers the gate; a reply with edits is folded in once by the approval agent.

Architecture

src/
├── main.rs          bootstrap: config, space, db, App, event loop
├── app/             the state machine (one module per feature)
│   ├── mod.rs       App struct, popups, settings, run_command, gate plumbing
│   ├── chat.rs      request lifecycle: history build, streaming, tool loop
│   ├── research.rs  the research pipeline: survey → plan → searchers → …
│   ├── swarm.rs     multi-persona roundtable
│   ├── watches.rs   standing research jobs
│   ├── sessions.rs, spaces.rs, models.rs, backends.rs, memory.rs
│   ├── files.rs, images.rs, scripts.rs, apps.rs   space artifacts
│   ├── skills_popup.rs, compaction.rs, export.rs, copy.rs, transcribe.rs
│   └── tests.rs     app-level integration tests
├── provider/
│   ├── mod.rs       message shapes, tool-call wire format, events
│   └── openrouter.rs  one client for all OpenAI-wire backends (OR/OpenAI/…)
├── tools.rs         the model's tools: search, fetch, python, video, apps…
├── tools (cont.)    skills.rs, extract.rs (PDF/OCR/text), citations.rs
├── db.rs            SQLite: sessions, messages, citations, model prefs
├── appserver.rs     localhost static server for model-created apps (port 8642)
├── ui/              ratatui rendering: history, popups, theme, markdown
├── input.rs         composer, @-autocomplete, command catalog
├── events.rs        key/mouse handling, clipboard
└── config.rs, space.rs   credentials, spaces layout

Data lives under the XDG data dir: spaces/<space>/ holds the per-space SQLite db, files, scripts, apps, and generated media.

Development

scripts/check.sh          # fmt + clippy (-D warnings, pedantic) + cargo-audit + tests
cargo test --bin nexus   # 428 tests, no network needed

Known limitations

• The research survey section is always visible, not collapsible.
• The survey's first-round questions are generated from the topic alone — the concurrent known-chunks/web-survey context arrives after round 1.
• A plan rework presented within the same second overwrites the plan file (same timestamped name).