@sinlair/reagent

ReAgent

Standalone local-first runtime and CLI for long-running research workflows.

Use one always-on runtime for research, memory, channels, and delivery surfaces.

English  |  简体中文

Roadmap  |  Operations  |  Contributing

At A Glance

• reagent is the primary product surface.
• the only official npm install target is @sinlair/reagent
• The runtime is designed to stay on and act as one control plane for research, memory, channels, and delivery.
• The CLI is the main control surface. The web console is a companion inspection surface.
• Research work is durable: briefs, tasks, reports, presentations, module assets, and file-backed memory stay visible after one run.

What ReAgent Is

ReAgent is not a chat wrapper with a few extra research commands.

It is a standalone runtime and CLI for people who want:

• structured research briefs instead of ad-hoc prompts
• durable memory instead of chat-only context
• visible long-running task state instead of hidden background work
• reusable outputs instead of one-off answers

ReAgent is currently strongest for:

• research task orchestration
• file-backed workspace memory
• evidence-backed reports
• reusable delivery artifacts
• local runtime and service control

Product Shape

The main product loop is:

• reagent onboard
• reagent home
• reagent service ...
• reagent runtime ...
• reagent research ...
• reagent memory ...
• optional web console inspection

The design principle is simple:

• one standalone runtime
• one CLI
• many durable artifacts

Quick Start

• Install dependencies.
• Push the Prisma schema.
• Start the runtime or onboard first.

npm install
npm run db:push
npm run dev

Minimal local setup:

LLM_PROVIDER=fallback
WECHAT_PROVIDER=mock

If PowerShell blocks npm, use npm.cmd.

Global CLI Install

npm install -g @sinlair/reagent
reagent onboard
reagent home
reagent service run

The published package name is @sinlair/reagent. The installed command is reagent.

Recommended CLI Flow

First-run flow:

reagent onboard
reagent home
reagent service run

Repair flow:

reagent doctor
reagent doctor --fix --skip-db

Daily control flow:

reagent home
reagent runtime status
reagent runtime logs --follow
reagent research recent
reagent research tasks
reagent memory recall "recent research choices"
reagent channels status

Example research flow:

reagent research enqueue "multimodal web agents" --question "Which open-source baselines are strongest?"
reagent research tasks
reagent research task <taskId>
reagent research report <taskId>

Runtime And Service Model

ReAgent is built around an always-on runtime.

Use:

• reagent home for the product overview
• reagent onboard for first-run setup
• reagent doctor for diagnostics and safe local repair
• reagent service ... to manage the supervised runtime
• reagent runtime ... to inspect health, status, dashboard, logs, and doctor output

Always-on service lifecycle:

reagent service install
reagent service status
reagent service start
reagent service restart
reagent service stop

Web Console

The web console is a companion surface, not the main control plane.

Use it for:

• artifact inspection
• graph and memory exploration
• task and report browsing
• operational visibility

Use the CLI for:

• first-run setup
• runtime control
• day-to-day automation
• scripting and repeatable operations

Knowledge And Artifact Model

ReAgent keeps research state durable.

Key layers:

• MEMORY.md for long-term memory
• memory/YYYY-MM-DD.md for daily notes
• indexed metadata in memory-index.json
• research task runs in research/task-runs.json
• research rounds in research/rounds/<taskId>/
• reports, reviews, presentations, and module assets as reusable artifacts

This is intentional:

• memory is not just chat history
• research is not just one synchronous run
• outputs are meant to be reopened and reused

Repository Layout

Path	Purpose
./	Root ReAgent runtime, CLI, web console, and API server
docs/	Product visuals and supporting docs

Run Modes

Mode	Commands	Notes
Development	npm run dev	Local development with live reload
PM2	npm run pm2:start / npm run pm2:restart / npm run pm2:logs	Background runtime with PM2
Windows Service	npm run service:install / npm run service:status / npm run service:start / npm run service:stop	Machine-level runtime on Windows

See OPERATIONS.md for deployment and maintenance details.

OpenClaw

OpenClaw is not the main product direction.

It is useful as:

• a reference for host/runtime design
• a compatibility path
• an external integration target

If you are adopting ReAgent, install the standalone package:

npm install -g @sinlair/reagent

ReAgent does not ship a second public adapter package anymore. The product install path is the standalone @sinlair/reagent CLI.

Development

Validation:

npm run check:all
npm run test

If you only want to validate the standalone publishable package:

npm run release:verify
npm run release:pack

Documentation

• Chinese README: README.zh-CN.md
• Roadmap: ROADMAP.md
• Operations: OPERATIONS.md
• Contributing: CONTRIBUTING.md
• Security: SECURITY.md
• Research agent landscape notes: docs/research-agent-landscape.md

Inspiration

ReAgent is informed by projects such as:

• GPT Researcher
• deer-flow
• PASA
• Paper2Agent
• enterprise-deep-research
• InternAgent
• OpenClaw

License

Released under the MIT License.