ResearchClaw is a personal AI assistant built for research: fast to set up, easy to run locally or in the cloud, and ready to integrate with the chat apps you already use. With extensible skills, it helps you streamline literature review, note-taking, experiment tracking, and paper writing—end to end.
See the code
Local-first Research OS for papers, workflows, experiments, channels, and automation.
English | 中文 | Docs | Roadmap | Research-Equality Ecosystem
Part of the Research-Equality ecosystem for AI-native research workflows.
Persistent research state · Multi-agent runtime · Skills + MCP · Automation + channels
Why ResearchClaw • Quick Start • Research-Equality Ecosystem • What You Get Today • Docs
ResearchClaw is the runtime and workspace layer of the broader Research-Equality stack. It keeps long-horizon research work durable: projects, workflows, claims, evidence, experiments, artifacts, reminders, channels, and automation all live in one local-first system instead of dissolving across chat threads, terminals, and scattered folders.
| Common AI research workflow pain | What ResearchClaw does instead |
|---|---|
| Research work disappears into one-off chats or shell history | Persists project -> workflow -> task -> artifact state with notes, claims, evidence, drafts, and reminders |
| Tools for search, execution, writing, and follow-up are split apart | Puts console, automation, channels, APIs, papers, experiments, and memory into one runtime |
| It is hard to hand work over between web, terminal, and messaging surfaces | Exposes the same research state through the web console, IM channels, cron jobs, sessions, and control-plane APIs |
| Skills, providers, and external tools are glued together ad hoc | Standardizes SKILL.md, MCP, provider routing, fallback chains, and per-agent workspace rules |
Under the hood, the current codebase combines:
console, telegram, discord, dingtalk, feishu, imessage, qq, and voiceSKILL.md support, Skills Hub search/install APIs, MCP client management, and custom channelsIt is still an Alpha project, but it is no longer just a platform shell. The code now includes a minimal research workflow runtime, claim/evidence graph, experiment tracking, blocker remediation, and project dashboard. The biggest remaining gaps are evidence-matrix quality, stronger claim-evidence validation, richer external execution adapters, and submission/reproducibility packaging.
git clone https://github.com/MingxinYang/ResearchClaw.git
cd ResearchClaw
pip install -e .
researchclaw init --defaults --accept-security
This creates:
~/.researchclaw~/.researchclaw.secretSOUL.md, AGENTS.md, PROFILE.md, and HEARTBEAT.mdresearchclaw models config
Or add one directly:
researchclaw models add openai --type openai --model gpt-5 --api-key sk-...
Supported provider types in code today: openai, anthropic, gemini, ollama, dashscope, deepseek, minimax, other, custom.
researchclaw app --host 127.0.0.1 --port 8088
Open http://127.0.0.1:8088.
If the page says Console not found, build the frontend once:
cd console
npm install
npm run build
The backend automatically serves console/dist when it exists.
After startup, open the Research page in the console to:
ResearchClaw works best as the persistent runtime and workspace in a larger skill ecosystem. The companion repositories below cover stage-specific research work, while the awesome repository maps the wider AI scientist and AI-for-research landscape.
Browse the full organization here: Research-Equality repositories
| Repository | Role next to ResearchClaw | Use it when |
|---|---|---|
| RE-idea-generation | authoritative skills for idea generation, problem discovery, and direction exploration | you need to turn vague interests into defensible research directions |
| RE-literature-discovery | authoritative skills for literature discovery, authority-aware ranking, evidence synthesis, and survey writing | you want auditable paper search, filtering, and review workflows |
| RE-research-design | authoritative skills for research design, method formalization, experiment planning, and evaluation design | you need a stronger design layer before implementation starts |
| RE-experiment | authoritative skills for experiment planning, implementation, validation, and analysis | you are reproducing baselines, running ablations, or tightening experiment traceability |
| RE-paper-writing | authoritative skills for paper planning, drafting, revision, LaTeX workflows, and submission QA | you want the writing and submission stack to stay connected to real artifacts |
| awesome-ai-scientists | the Awesome-AI-Research landscape map for AI-native research systems, workflow modules, benchmarks, surveys, datasets, and meta-resources | you want a broader map of AI scientist systems and AI research tooling beyond this project |
A practical pairing is ResearchClaw plus one or two RE-* repositories for the stage you are actively pushing, with awesome-ai-scientists as the discovery layer for adjacent tools, systems, and benchmarks.
/api/health, /api/version, /api/control/*, /api/automation/*, /api/providers, /api/skills, /api/mcp, /api/workspace, and moreproject -> workflow -> task -> artifact relationshipsliterature_search, paper_reading, note_synthesis, hypothesis_queue, experiment_plan, experiment_run, result_analysis, writing_tasks, and review_and_followupBuilt-in tools registered by the agent include:
semantic_scholar_searchbibtex_search, bibtex_add_entry, bibtex_exportlatex_template, latex_compile_checkdata_describe, data_queryrun_shell, read_file, write_file, edit_file, append_filebrowse_url, browser_use, send_file, memory_searchskills_list, skills_activate, skills_read_fileBundled skills currently shipped in src/researchclaw/agents/skills/ include:
arxivbrowser_visiblecitation_networkcrondingtalk_channeldocxexperiment_trackerfigure_generatorfile_readerhimalayaliterature_reviewnewspaper_summarizerpdfpptxresearch_notesresearch_workflowsxlsxRuntime data lives under the working directory, while secrets are stored separately:
~/.researchclaw/
├── config.json
├── jobs.json
├── chats.json
├── research/
│ └── state.json
├── sessions/
├── active_skills/
├── customized_skills/
├── papers/
├── references/
├── experiments/
├── memory/
├── md_files/
├── custom_channels/
└── researchclaw.log
~/.researchclaw.secret/
├── envs.json
└── providers.json
Provider credentials and persisted environment variables are intentionally kept out of the working directory.
Backend checks:
pip install -e ".[dev]"
PYTHONPATH=src pytest -q
Console build:
npm --prefix console run build
Website build:
corepack pnpm --dir website run build
Repo-wide helper:
scripts/check-ci.sh --skip-install
Main documentation files in this repository:
The current codebase is best described as:
ResearchClaw is a personal AI assistant built for research: fast to set up, easy to run locally or in the cloud, and ready to integrate with the chat apps you already use. With extensible skills, it helps you streamline literature review, note-taking, experiment tracking, and paper writing—end to end.
See the code
Local-first Research OS for papers, workflows, experiments, channels, and automation.
English | 中文 | Docs | Roadmap | Research-Equality Ecosystem
Part of the Research-Equality ecosystem for AI-native research workflows.
Persistent research state · Multi-agent runtime · Skills + MCP · Automation + channels
Why ResearchClaw • Quick Start • Research-Equality Ecosystem • What You Get Today • Docs
ResearchClaw is the runtime and workspace layer of the broader Research-Equality stack. It keeps long-horizon research work durable: projects, workflows, claims, evidence, experiments, artifacts, reminders, channels, and automation all live in one local-first system instead of dissolving across chat threads, terminals, and scattered folders.
| Common AI research workflow pain | What ResearchClaw does instead |
|---|---|
| Research work disappears into one-off chats or shell history | Persists project -> workflow -> task -> artifact state with notes, claims, evidence, drafts, and reminders |
| Tools for search, execution, writing, and follow-up are split apart | Puts console, automation, channels, APIs, papers, experiments, and memory into one runtime |
| It is hard to hand work over between web, terminal, and messaging surfaces | Exposes the same research state through the web console, IM channels, cron jobs, sessions, and control-plane APIs |
| Skills, providers, and external tools are glued together ad hoc | Standardizes SKILL.md, MCP, provider routing, fallback chains, and per-agent workspace rules |
Under the hood, the current codebase combines:
console, telegram, discord, dingtalk, feishu, imessage, qq, and voiceSKILL.md support, Skills Hub search/install APIs, MCP client management, and custom channelsIt is still an Alpha project, but it is no longer just a platform shell. The code now includes a minimal research workflow runtime, claim/evidence graph, experiment tracking, blocker remediation, and project dashboard. The biggest remaining gaps are evidence-matrix quality, stronger claim-evidence validation, richer external execution adapters, and submission/reproducibility packaging.
git clone https://github.com/MingxinYang/ResearchClaw.git
cd ResearchClaw
pip install -e .
researchclaw init --defaults --accept-security
This creates:
~/.researchclaw~/.researchclaw.secretSOUL.md, AGENTS.md, PROFILE.md, and HEARTBEAT.mdresearchclaw models config
Or add one directly:
researchclaw models add openai --type openai --model gpt-5 --api-key sk-...
Supported provider types in code today: openai, anthropic, gemini, ollama, dashscope, deepseek, minimax, other, custom.
researchclaw app --host 127.0.0.1 --port 8088
Open http://127.0.0.1:8088.
If the page says Console not found, build the frontend once:
cd console
npm install
npm run build
The backend automatically serves console/dist when it exists.
After startup, open the Research page in the console to:
ResearchClaw works best as the persistent runtime and workspace in a larger skill ecosystem. The companion repositories below cover stage-specific research work, while the awesome repository maps the wider AI scientist and AI-for-research landscape.
Browse the full organization here: Research-Equality repositories
| Repository | Role next to ResearchClaw | Use it when |
|---|---|---|
| RE-idea-generation | authoritative skills for idea generation, problem discovery, and direction exploration | you need to turn vague interests into defensible research directions |
| RE-literature-discovery | authoritative skills for literature discovery, authority-aware ranking, evidence synthesis, and survey writing | you want auditable paper search, filtering, and review workflows |
| RE-research-design | authoritative skills for research design, method formalization, experiment planning, and evaluation design | you need a stronger design layer before implementation starts |
| RE-experiment | authoritative skills for experiment planning, implementation, validation, and analysis | you are reproducing baselines, running ablations, or tightening experiment traceability |
| RE-paper-writing | authoritative skills for paper planning, drafting, revision, LaTeX workflows, and submission QA | you want the writing and submission stack to stay connected to real artifacts |
| awesome-ai-scientists | the Awesome-AI-Research landscape map for AI-native research systems, workflow modules, benchmarks, surveys, datasets, and meta-resources | you want a broader map of AI scientist systems and AI research tooling beyond this project |
A practical pairing is ResearchClaw plus one or two RE-* repositories for the stage you are actively pushing, with awesome-ai-scientists as the discovery layer for adjacent tools, systems, and benchmarks.
/api/health, /api/version, /api/control/*, /api/automation/*, /api/providers, /api/skills, /api/mcp, /api/workspace, and moreproject -> workflow -> task -> artifact relationshipsliterature_search, paper_reading, note_synthesis, hypothesis_queue, experiment_plan, experiment_run, result_analysis, writing_tasks, and review_and_followupBuilt-in tools registered by the agent include:
semantic_scholar_searchbibtex_search, bibtex_add_entry, bibtex_exportlatex_template, latex_compile_checkdata_describe, data_queryrun_shell, read_file, write_file, edit_file, append_filebrowse_url, browser_use, send_file, memory_searchskills_list, skills_activate, skills_read_fileBundled skills currently shipped in src/researchclaw/agents/skills/ include:
arxivbrowser_visiblecitation_networkcrondingtalk_channeldocxexperiment_trackerfigure_generatorfile_readerhimalayaliterature_reviewnewspaper_summarizerpdfpptxresearch_notesresearch_workflowsxlsxRuntime data lives under the working directory, while secrets are stored separately:
~/.researchclaw/
├── config.json
├── jobs.json
├── chats.json
├── research/
│ └── state.json
├── sessions/
├── active_skills/
├── customized_skills/
├── papers/
├── references/
├── experiments/
├── memory/
├── md_files/
├── custom_channels/
└── researchclaw.log
~/.researchclaw.secret/
├── envs.json
└── providers.json
Provider credentials and persisted environment variables are intentionally kept out of the working directory.
Backend checks:
pip install -e ".[dev]"
PYTHONPATH=src pytest -q
Console build:
npm --prefix console run build
Website build:
corepack pnpm --dir website run build
Repo-wide helper:
scripts/check-ci.sh --skip-install
Main documentation files in this repository:
The current codebase is best described as: