mirror of
https://github.com/getcompanion-ai/alpha-hub.git
synced 2026-04-15 03:00:43 +00:00
081d32d50c
- Add `alpha status` command to check auth state - Search results drop raw text blobs by default (opt-in via includeRaw) - Clean and normalize search result fields - Export parsePaperSearchResults for external consumers - Login callback server handles EADDRINUSE, styled HTML pages - Better browser open error handling on all platforms - Search prints mode description before results Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
113 lines
4.3 KiB
Markdown
113 lines
4.3 KiB
Markdown
# Alpha Hub
|
|
|
|
Research agents hallucinate paper details and forget what they learn in a session. Alpha Hub gives them semantic paper search, AI-generated reports, and persistent annotations — so they get smarter with every task. Search and content powered by [alphaXiv](https://alphaxiv.org).
|
|
|
|
[](LICENSE)
|
|
[](https://nodejs.org)
|
|
|
|
## Quick Start
|
|
|
|
```bash
|
|
npm install -g @companion-ai/alpha-hub
|
|
alpha login # sign in with alphaXiv
|
|
alpha status # show whether alphaXiv auth is present
|
|
alpha search "attention mechanism" # search papers
|
|
alpha get 1706.03762 # fetch paper report
|
|
```
|
|
|
|
## How It Works
|
|
|
|
Alpha is designed for your coding agent to use (not for you to use!). You can prompt your agent to use it (e.g., "Use the CLI command `alpha` to search for papers on LoRA. Run `alpha` to see how it works.")
|
|
|
|
**Most of the time, it's simple — search, read, use:**
|
|
|
|
```bash
|
|
alpha search "transformer attention" # find relevant papers
|
|
alpha get 1706.03762 # fetch AI-generated paper report
|
|
# Agent reads the report, understands the paper. Done.
|
|
```
|
|
|
|
**When the agent discovers something useful**, it can annotate locally for next time:
|
|
|
|
```bash
|
|
alpha annotate 1706.03762 "Superseded by Flash Attention for efficiency"
|
|
|
|
# Next session, the annotation appears automatically on alpha get.
|
|
```
|
|
|
|
**Need to go deeper?** Ask questions about any paper:
|
|
|
|
```bash
|
|
alpha ask 1706.03762 "What datasets were used for evaluation?"
|
|
```
|
|
|
|
## Commands
|
|
|
|
| Command | Purpose |
|
|
|---------|---------|
|
|
| `alpha search <query>` | Search papers (semantic, keyword, or agentic) |
|
|
| `alpha get <id\|url>` | Fetch paper report + local annotation |
|
|
| `alpha ask <id\|url> <question>` | Ask a question about a paper |
|
|
| `alpha code <github-url> [path]` | Read files from a paper repository |
|
|
| `alpha annotate <id> <note>` | Attach a note to a paper |
|
|
| `alpha annotate <id> --clear` | Remove a note |
|
|
| `alpha annotate --list` | List all notes |
|
|
| `alpha login` | Sign in with alphaXiv |
|
|
| `alpha status` | Show alphaXiv authentication status |
|
|
| `alpha logout` | Sign out |
|
|
|
|
All commands accept `--json` for machine-readable output.
|
|
|
|
## Self-Improving Agents
|
|
|
|
Alpha Hub is designed for a loop where agents get better over time.
|
|
|
|
**Annotations** are local notes that agents attach to papers. They persist across sessions and appear automatically on future fetches — so agents learn from past experience.
|
|
|
|
```
|
|
Without Alpha Hub With Alpha Hub
|
|
───────────────── ──────────────
|
|
Search the web for papers Semantic search via alphaXiv
|
|
Read raw PDFs AI-generated paper reports
|
|
Miss context and gotchas Agent notes what it learns
|
|
Knowledge forgotten ↗ Even smarter next session
|
|
↻ Repeat next session
|
|
```
|
|
|
|
## Key Features
|
|
|
|
### Semantic Search
|
|
|
|
Three search modes — semantic (embedding similarity), keyword (exact terms), and agentic (multi-turn retrieval) — so agents find the right papers regardless of how they phrase the query. `--mode all` runs all three in parallel for maximum recall.
|
|
|
|
```bash
|
|
alpha search "methods for reducing hallucination in LLMs" # semantic
|
|
alpha search "LoRA" --mode keyword # keyword
|
|
alpha search "retrieval-augmented generation for QA" --mode agentic
|
|
alpha search "alignment of vision language models" --mode all
|
|
```
|
|
|
|
### Paper Q&A
|
|
|
|
Ask questions about any paper without reading the full PDF. The answer is grounded in the paper's actual content.
|
|
|
|
```bash
|
|
alpha ask 2106.09685 "What is the rank used for the low-rank matrices?"
|
|
```
|
|
|
|
### Annotations
|
|
|
|
Local notes that agents attach to papers — they persist across sessions and appear automatically on future fetches. See the annotation as a gap the agent discovered and recorded so it doesn't repeat the same mistake.
|
|
|
|
### Repository Reading
|
|
|
|
Read files directly from a paper's GitHub repository when the implementation matters.
|
|
|
|
```bash
|
|
alpha code https://github.com/openai/gpt-2 /
|
|
alpha code https://github.com/openai/gpt-2 src/model.py
|
|
```
|
|
|
|
## License
|
|
|
|
[MIT](LICENSE)
|