StepFun (阶跃星辰) - Setup Guide¶
Source: fengjunlu618/stepfun-skills Category: LLM Provider / Multimodal
The first Chinese LLM provider skill suite for Hermes Agent. Provides access to StepFun's full model lineup - multimodal reasoning, text-to-speech, speech recognition, image generation, real-time voice, and search/RAG - all through a Step Plan subscription channel (separate from standard API billing).
Installation¶
# Clone the repo
git clone https://github.com/fengjunlu618/stepfun-skills.git
mkdir -p ~/.hermes/skills/llm-providers/stepfun
cp -a stepfun-skills/* ~/.hermes/skills/llm-providers/stepfun/
Individual skills:
| Skill | Command |
|---|---|
| stepfun-chat | Chat Completions - text reasoning, multimodal, Tool Call, routing |
| stepfun-tts | TTS + voice cloning - Contextual TTS, 19 official voices |
| stepfun-image | Text-to-image + image editing - 1-2s generation |
| stepfun-asr | Speech recognition - async file + streaming |
| stepfun-realtime | Real-time voice conversation - speech ↔ speech |
Prerequisites¶
| Requirement | Details |
|---|---|
| StepFun Account | Register at platform.stepfun.com |
| Step Plan Subscription | Required - calls go through Step Plan (not standard API credit) |
| STEPFUN_API_KEY | Set in ~/.hermes/.env: STEPFUN_API_KEY=sk-... |
| Hermes Agent | v0.20.0+ |
Key Capabilities¶
Model Lineup¶
| Model | Type | Price | Capability |
|---|---|---|---|
step-3.7-flash |
Multimodal Reasoning | ¥1.35/1M tokens | Native multimodal (image/video) + reasoning + agent + Tool Call |
step-3.5-flash-2603 |
Agent Optimized | ¥0.7/1M tokens | Reasoning + agent + low-reasoning mode |
step-3.5-flash |
Flagship Reasoning | ¥0.7/1M tokens | Reasoning + Tool Call |
step-router-v1 |
Smart Router | Per-hit billing | Pro + Flash dual engine |
step-1o-turbo-vision |
Vision Understanding | ¥2.5/1M tokens | Image/video understanding |
stepaudio-2.5-tts |
Speech Synthesis | ¥5.8/10K chars | Contextual TTS |
step-tts-mini |
Speech Synthesis | ¥0.9/10K chars | CN/EN/JP/Cantonese/Sichuan |
stepaudio-2.5-asr |
Speech Recognition | ¥0.15/hour | 4B MTP ultra-fast |
stepaudio-2.5-realtime |
Real-time Voice | ¥10/1M tokens | Speech ↔ speech |
step-image-edit-2 |
Image Gen/Edit | ¥0.02/image | Text-to-image + editing |
step-2x-large |
Text-to-Image | ¥0.1/image | Strong text rendering |
Core Features¶
| Capability | How to Trigger | Notes |
|---|---|---|
| Chat/Multimodal | Load stepfun-chat skill, use natural language |
Supports image + video input |
| Text-to-Speech | Load stepfun-tts skill, "Generate speech from [text]" |
19 official voices + voice cloning |
| Image Generation | Load stepfun-image skill, "Generate image of [description]" |
1-2s generation, text-in-image support |
| Speech Recognition | Load stepfun-asr skill, "Transcribe [audio file]" |
Async file + streaming modes |
| Real-time Voice | Load stepfun-realtime skill |
Speech ↔ speech conversation |
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Multilingual Content | Use StepFun for Chinese market content generation and analysis |
| Voice Agent Pipeline | ASR → Chat → TTS for autonomous phone/voice operations |
| Cost Optimization | Route Chinese-language tasks to StepFun (¥0.7/1M tokens vs GPT-4 pricing) |
| Video Analysis | step-1o-turbo-vision for Chinese video content understanding |
| Image Generation | step-image-edit-2 for social media assets at ¥0.02/image |
Troubleshooting¶
| Issue | Fix |
|---|---|
| 401 Unauthorized | Verify STEPFUN_API_KEY in ~/.hermes/.env |
| Billing errors | Ensure Step Plan subscription is active (not standard API credit) |
| Model not found | Check model name exactly - StepFun model IDs are case-sensitive |
| Rate limiting | Step Plan has per-minute quotas; stagger requests |
Verification¶
# Verify skill installed
hermes skills list | grep stepfun
# Test API key
curl -s -H "Authorization: Bearer $STEPFUN_API_KEY" \
"https://api.stepfun.com/v1/models" | head -20
← Skills Catalog | Discovery Page → Powered by CorpusIQ
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.