Guide
Hermes Agent + Paperclip AI: Wire Paperclip File Tasks to a Hermes Backend
Paperclip AI handles document-grounded tasks — summarize this PDF, extract this CSV, label these images. Hermes Agent runs the long-running agent layer that decides when to call Paperclip, where to put the results, and which channel (Telegram, Discord, web) to reply on.
What Paperclip AI Does Well
- Multi-format document ingestion (PDF, DOCX, XLSX, CSV, images)
- Structured extraction (table-to-JSON, form fields, OCR)
- Long-doc summarization with citations
What Hermes Does Well
- Stateful agent loops — remembers what you uploaded yesterday
- Multi-channel reach — the same agent fields requests on Telegram, Discord, web
- Tool calls — can chain Paperclip extraction into a Slack message, a DB write, a follow-up email
Architecture
Telegram / Discord / Web → Hermes Agent
↓ tool_call: paperclip.extract
Paperclip API
↓ structured JSON
Hermes Agent → reply / store / forwardStep 1: Get a Paperclip API Key
Sign up at paperclip.ai, generate an API key in the dashboard.
Step 2: Add Paperclip as a Hermes Skill
Create a skill at ~/.hermes/skills/paperclip-extract/SKILL.md:
---
name: paperclip-extract
description: Send a file URL to Paperclip AI for structured extraction. Use when user uploads a PDF/DOCX/XLSX and asks to extract / summarize / parse.
---
# Paperclip Extract
Pipe a file through Paperclip AI's extraction endpoint.
## Steps
1. Get the file URL from the user message or attached file
2. Call `scripts/extract.sh <url> <prompt>`
3. Return the JSON output formatted for the channel
And the script:
#!/bin/bash
# ~/.hermes/skills/paperclip-extract/scripts/extract.sh
curl -s -X POST https://api.paperclip.ai/v1/extract \
-H "Authorization: Bearer $PAPERCLIP_API_KEY" \
-H "Content-Type: application/json" \
-d "{\"url\": \"$1\", \"prompt\": \"$2\"}"Step 3: Pass the Key to Hermes
docker run -d \
-e PAPERCLIP_API_KEY="pk_..." \
-v $HOME/.hermes:/root/.hermes \
ghcr.io/nousresearch/hermes-agent:latestStep 4: Test From a Channel
Send a Telegram message: “Summarize this PDF: https://example.com/report.pdf”. Hermes routes to the paperclip-extract skill, which posts the URL to Paperclip, gets the structured response, and replies in Telegram.
Hermes on OpenClaw Launch + Paperclip
For managed Hermes, drop the same skill into the Skills Hub installer. The Hermes Agent Skills guide covers Dashboard-driven skill installation. Set PAPERCLIP_API_KEY in the env tab of your instance.
Why Not Use Paperclip Standalone?
Paperclip is request/response — you give it a file, it gives you JSON. It doesn't remember previous extractions, doesn't talk on Telegram, doesn't chain calls. Hermes provides the agent loop around Paperclip's file-processing primitive.
Limits
- Paperclip rate-limits at 100 req/min on the free tier — queue Hermes calls if you expect bursts.
- Max file size: 50MB per request — split larger docs before extraction.
- Cold-start latency for first request after idle — Hermes should fall back gracefully on 503.
What's Next?
- OpenClaw Launch vs Paperclip AI — the platform-level comparison
- Hermes Agent Skills — skill protocol details
- Hermes Agent + MCP — alternative to skills for tool calls
- Hermes Agent BYOK — pick an LLM for the agent layer