A self-hostable AI research workspace for grounded chat, paper study, scientific skills, and research execution.
Grounded over your files. Structured around papers. Ready for execution.
English · 简体中文 · 日本語 · Français · Deutsch
Documentation · Quick Start · Community
Scan to join our community · 扫码加入飞书/微信体验群
InnoClaw turns server-side folders into AI-native workspaces for grounded chat, paper study, scientific workflows, and research execution.
It is built for researchers, developers, labs, and self-hosters who want more than a generic chat UI: cited answers over real files, reusable skills, and a path from reading to execution.
- ClawHub Skill Import: New integration to import skills directly from ClawHub via a dedicated API endpoint and import dialog
- Code Preview Panel: New in-editor code preview component supporting syntax highlighting and save-status tracking
- Paper Study Cache: Persistent caching layer for paper study sessions, improving reload performance and state continuity
- Multimodal Vision for Paper Analysis: PDF images are now extracted and analyzed visually during paper discussion and research ideation sessions
- Claude Code Skills Integration: Import skills directly from local folders or Claude Code projects via a new dedicated import workflow
- Multimodal Vision for Paper Discussion & Ideation: Vision-capable providers can now receive extracted PDF page images alongside text so discussion and ideation agents can analyze figures, tables, and diagrams.
- Paper Pages Gallery UI: Discussion and ideation panels now show a collapsible thumbnail gallery for extracted paper pages with full-size preview dialogs.
- Provider Vision Capability Detection: Provider configs now expose vision support so routes can switch between multimodal and text-only paper context automatically.
- Remote Job Profile Management & SSH Hardening: Secure remote profile creation, editing, and SSH-hardened job submission for research execution
- Rich Markdown Rendering in Agent Panel: Agent messages now render tables, LaTeX math, and syntax-highlighted code blocks
- API Provider Settings UI: Configure AI provider API keys and endpoints directly from the Settings page
- rjob Profile Config & Submission Hardening: Remote profiles now store full rjob defaults (image, GPU, CPU, memory, mounts, charged-group, private-machine, env vars, host-network, example commands).
submitRemoteJobbuilds the rjob command internally from stored config - the agent can no longer modify flags like--charged-groupor--image. SSH transport fixed with-o StrictHostKeyChecking=no -tt, init script sourcing, and double-quote wrapping for correct quoting. - Profile Editing: Edit button (pencil icon) on remote profiles in the Remotes tab. Click to load profile into the form for updating, including all rjob config fields.
- Direct Job Submission Shortcut: Agent-Long mode can skip inspect/patch/sync stages for simple job submissions:
listRemoteProfiles -> prepareJobSubmission -> approval -> submitRemoteJob.
- Paper Discussion & Ideation Robustness: Per-role token budgets (2-2.5x increase), automatic retry on empty/short responses, and error visibility in the UI. Fixes agents returning empty or truncated output with reasoning-capable models (SH-Lab, Qwen, etc.)
- Full Paper Context: Discussion and ideation agents now receive up to 30k chars of the full paper text (local files) instead of just the abstract, enabling deeper analysis of methodology, experiments, and results
- Abstract Extraction Fix: Heuristic regex-based abstract extraction with improved AI prompt to prevent extracting author names instead of the actual abstract
- Research Execution Engine: New AI-driven research orchestration system with remote profiles, capability toggles, run history, and agent tools
- Auto-updating README "What's New": GitHub Actions workflow that automatically generates and commits a What's New section daily
No entries yet. This section is auto-updated when significant new features are detected by CI.
InnoClaw is a self-hostable web app for research-centric knowledge work. It combines workspace management, retrieval-augmented chat, paper search and review, reusable scientific skills, and agent-based execution in one place.
Instead of juggling separate tools for files, notes, literature review, and automation, you keep the workflow in one workspace: open a folder, sync content, ask grounded questions, study papers, and run multi-step research tasks.
- Workspace-first - Treat server folders as durable research workspaces with files, notes, chat history, and execution context
- Grounded AI answers - Use RAG-backed chat with source citations over your own documents and code
- Research-native workflows - Study papers, run structured multi-agent discussions, and generate new directions from literature
- Scientific skills built in - Import and use 206 SCP scientific skills across domains such as drug discovery, genomics, and protein science
- Execution, not just conversation - Move from reading and planning to job submission, monitoring, result collection, and next-step recommendations
- Self-hosted and multi-model friendly - Run with OpenAI, Anthropic, Gemini, and compatible endpoints in your own environment
git clone https://github.com/SpectrAI-Initiative/InnoClaw.git
cd InnoClaw
npm install
npm run dev- Open
http://localhost:3000 - Configure one AI provider from the Settings page
- Open or clone a workspace, then click
Syncto build the RAG index - Need OS-specific prerequisites or production setup? See
docs/getting-started/installation.md
- Chat with local files and code using grounded citations
- Search, summarize, and review papers in one workspace
- Run 5-role structured paper discussions for critique and reproducibility thinking
- Generate summaries, FAQs, briefs, timelines, and research ideas
- Import scientific skills and trigger reusable domain workflows
- Manage remote research tasks with approval gates, monitoring, and result analysis
| If you want to... | Start here | What happens next |
|---|---|---|
| Chat with your own files | Workspace + RAG Chat | Open a folder, click Sync, and ask cited questions |
| Read and break down papers | Paper Study | Search papers, summarize them, then move into discussion or notes |
| Stress-test ideas with multiple perspectives | Multi-Agent Discussion | Run role-based reviews for critique, evidence gathering, and reproducibility thinking |
| Turn reading into new directions | Research Ideation | Generate directions, compare options, and save outputs into notes |
| Execute research work on remote infrastructure | Research Execution Workspace | Review code, approve changes, submit jobs, monitor runs, and collect results |
| Layer | Role in the workflow |
|---|---|
| Workspace | Holds files, notes, session context, and project state |
| Knowledge | Syncs files into the RAG index so answers stay grounded |
| Paper Workbench | Handles literature search, summary, discussion, and ideation |
| Skills | Adds reusable domain workflows and tool-guided capabilities |
| Execution | Extends the workflow into remote jobs and experiment loops |
Search literature, preview papers, summarize them, and move directly into discussion or ideation.
- Search across multiple sources from one UI
- Use AI-assisted query expansion for broader coverage
- Open paper previews without leaving workspace context
- Save outputs into notes for reuse
Run a structured paper review with roles such as moderator, librarian, skeptic, reproducer, and scribe.
- Follow a deterministic staged discussion flow
- Compare evidence, methods, limitations, and reproducibility concerns
- Generate review records that are easier to scan than free-form chats
- Use full-paper context for deeper analysis
Go from code inspection to job submission and result analysis inside a guided execution workflow.
- Review repositories and propose patches with agent assistance
- Gate high-risk steps with explicit approval checkpoints
- Submit jobs through Shell, Slurm, or
rjobbackends - Monitor status, collect artifacts, and generate recommendations for the next step
| Feature | What it enables |
|---|---|
| Workspace Management | Map server folders into persistent AI workspaces |
| File Browser | Browse, upload, create, edit, preview, and sync files |
| RAG Chat | Ask grounded questions over indexed files with citations |
| Paper Study | Search, summarize, and inspect papers in one place |
| Discussion Mode | Run structured multi-role paper discussions |
| Research Ideation | Generate new directions and cross-disciplinary ideas |
| Skills System | Import reusable scientific and workflow skills |
| Research Execution | Orchestrate remote experiment loops with monitoring and approval gates |
| Multi-Agent Sessions | Keep separate execution contexts across tabs and projects |
| Multi-LLM Support | Use OpenAI, Anthropic, Gemini, and compatible endpoints |
- Start here - Overview, Installation
- Configure and deploy - Deployment, Environment Variables, Configuration
- Use the product - Features, API Reference
- Troubleshoot and contribute - Troubleshooting, Development Guide
- Need setup or usage help? Start with the docs at https://SpectrAI-Initiative.github.io/InnoClaw/
- Found a bug or want a feature? Open an issue at https://github.com/SpectrAI-Initiative/InnoClaw/issues
- Want direct discussion? Join the Feishu community from
docs/README_CN.md
- License - Apache-2.0, see
LICENSE - Repository - https://github.com/SpectrAI-Initiative/InnoClaw
- Docs - https://SpectrAI-Initiative.github.io/InnoClaw/