@sinlair/reagent

ReAgent

Local-first runtime and CLI for long-running research workflows.

Run research, memory, channels, jobs, and durable artifacts from one always-on workspace.

English  |  简体中文

Docs  |  Quickstart  |  Roadmap  |  Contributing

Why ReAgent

ReAgent is built for people who want research work to stay inspectable after the first run.

• Write structured briefs instead of ad-hoc prompts.
• Keep durable memory instead of relying only on chat history.
• Inspect jobs, hooks, sessions, and artifacts instead of trusting hidden orchestration.
• Reopen reports, reviews, presentations, and workstreams without recreating them.

Quickstart

Install the public package:

npm install -g @sinlair/reagent

Run the first-start flow:

reagent onboard
reagent onboard --apply
reagent home
reagent service run

Default web console URL:

http://127.0.0.1:3000/

The safe starter profile uses fallback + mock. It is good for evaluation and walkthroughs, not final research-quality output.

Typical Flow

Queue a research task, inspect progress, then reopen the durable outputs:

reagent research directions
reagent research enqueue "multimodal web agents" --question "Which open-source baselines are strongest?"
reagent research tasks
reagent research report <taskId>
reagent research handoff <taskId>
reagent memory recall "recent research choices"

What You Get

• One always-on runtime for research, memory, channels, jobs, and delivery artifacts.
• A CLI-first control plane, with the web console as an inspection surface.
• Durable artifacts such as briefs, tasks, reports, reviews, presentations, workstreams, and module assets.
• Local configuration for LLM routes, MCP servers, workspace skills, and inbound command policy.
• Optional WeChat and OpenClaw bridge surfaces when you need delivery and host-state inspection.

Product Surfaces

ReAgent currently centers on five practical surfaces:

• reagent research ... queue work, inspect task state, reopen artifacts, and drive discovery, synthesis, and handoff
• reagent memory ... write durable notes, search them, and run manual or scheduled compaction
• reagent agent ... inspect runtime sessions, hooks, profiles, history, host linkage, and bounded delegation
• reagent channels ... receive and push messages through mock, native, or openclaw providers
• reagent workspace ... manage snapshots, restore previews, support bundles, and local config state

Repository Layout

These directories do different jobs:

• src/ main ReAgent runtime, CLI, HTTP routes, providers, and services
• tests/ root regression and integration coverage for the shipped product surface
• docs/ user-facing docs, walkthroughs, release notes, and positioning
• web/ companion web console assets
• package/ companion OpenClaw Weixin package; not the main ReAgent install target
• upstream/openclaw/ imported upstream reference snapshot used for compatibility and parity work; not the main contribution surface for ReAgent features

Docs

Start here:

• Docs Hub
• Public Install And Quickstart
• Core Product Flows
• Public Walkthroughs
• OpenClaw Host Surface
• Positioning: ReAgent vs Hermes Agent
• Developer Extensions
• Operations

Positioning

ReAgent is a research-first workspace, not a thin chat wrapper and not a general-purpose agent shell.

It borrows runtime ideas from projects such as Hermes Agent and OpenClaw:

• Hermes Agent informs runtime seams such as hooks, tool orchestration, skills, jobs, and constrained delegation.
• OpenClaw informs bridge and host-surface compatibility where channel delivery is needed.

The core product value is still durable research work, durable memory, and reusable artifacts.

Inspiration

ReAgent is informed by projects such as:

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

License

Released under the MIT License.