Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.siliconflow.com/llms.txt

Use this file to discover all available pages before exploring further.

Already connected SiliconFlow to Hermes Agent? This guide takes you further, from choosing the right model on SiliconFlow to building your daily assistant on Discord. Plus, a set of practical tips and best practices to help you get more out of every session. New to the setup? Start with the v1 guide first to connect your Hermes Agent with SiliconFlow.

​
What’s New

The v1 guide covered the step-by-step process of integrating SiliconFlow APIs to Hermes Agent. Since then, Hermes Agent has grown significantly — now ranked #1 in OpenRouter’s Coding Agents Rankings — and introduced more powerful new features worth exploring. This guide will further focus on:
  • Connecting Hermes Agent to Discord as a daily assistant
  • Tips & best practices for getting the most out of your Hermes Agent
  • Choosing the right SiliconFlow model for your use case

​
Connect Hermes Agent to Discord

With SiliconFlow configured in Hermes Agent, you can deploy it as a Discord bot and start chatting with your AI assistant directly in Discord via DMs or server channels.

​
Step 1: Create a Discord Application

  1. Go to the Discord Developer Portal and sign in
  2. Click New Application
  3. Enter a name (e.g., Hermes Agent (SiliconFlow API)) and click Create
Newapplication Copy (3)

​
Step 2: Configure the Bot

  1. In the left sidebar, click Bot
  2. Reset and Copy Bot Token
    1. Click Reset Token.
    2. ⚠️Copy the token and store it securely. You will need it during the onboarding process.
  3. Under Authorization Flow,
    1. Set Public Bot to ON
    2. Leave Require OAuth2 Code Grant set to OFF
  4. Scroll down to Privileged Gateway Intents and enable:
    1. Server Members Intent
    2. Message Content Intent
    3. Presence Intent (optional)
  5. Click Save Changes
Bot Copy Copy (2)
Image (25) Copy (2)

​
Step 3: Generate the Invite URL

  1. In the left sidebar, click Installation
  2. Under Installation Contexts, enable Guild Install
  3. Set Install Link to Discord Provided Link
  4. Under Default Install Settings → Guild Install, select:
    1. Scopes: bot and applications.commands
    2. Permissions: View Channels, Send Messages, Read Message History, Attach Files, etc.
Image (26) Copy (1) Copy (2)
Image (27) Copy (2)

​
Step 4: Invite the Bot to Your Server

  1. Open the invite URL in your browser.
  2. In the Add to Server dropdown, select your server.
  3. Click Continue, then Authorize.
  4. Complete the CAPTCHA if prompted.
Image (35)

​
Step 5: Find Your Discord User ID

Hermes Agent uses your User ID to control who can interact with the bot.
  1. In Discord, go to Settings and toggle Developer Mode ON
  2. Right-click your username → Copy User ID
微信图片 20260522142824 1223 10

​
Step 6: Start the Gateway

  1. Run the setup in your terminal:
hermes gateway setup
  1. Select Discord when prompted, then paste your Bot Token and User ID.
Image (29) Copy (2)
Once complete, send the bot a message in Discord to verify your SiliconFlow-powered assistant is live.
For advanced configuration options (free-response channels, session isolation, slash commands, etc.), refer to the Hermes Agent Discord documentation.

​
Use case: Daily News Report

Tell Hermes bot which website you don’t want to miss, and it’ll check them on a schedule and deliver updates right to your Discord channel. Example:
“Follow the latest news on siliconflow.com every day and report the updates”
Hermes visits the site daily and posts a summary — no manual checking needed.
B7dc3f6d3c79f1e4d7911c503b83a068 (1)

​
Tips & Best Practices

​
Starting Out: Be Specific, Not Vague

Tell Hermes exactly what you need — the more detail, the better the result.
❌ help me with models selection ✅ I need to call a text generation API on SiliconFlow with at least 205K context, cache support, and under $0.5 per million input tokens — what are my options?
When you spell out your requirements — context length, pricing budget, feature needs — Hermes narrows down the options for you in one shot. A vague prompt means back-and-forth clarifications; a specific one means you get what you need right away.

​
Leveling Up: Run Tasks in Parallel, Stop Waiting in Line

Most people use Hermes one task at a time. But Hermes can spin up multiple sub-agents that work simultaneously — each with its own context — and only the final summaries come back to you. For example, instead of looking things up one by one:
“Check these three things in parallel: (1) which text generation models are available with json mode support on SiliconFlow, (2) how the image generation API request format works for SiliconFlow, (3) what the rate limits are for my SiliconFlow API tier.”
One message, three independent lookups, one consolidated answer. It’s faster and saves tokens because only the key findings are returned to your main conversation.

​
Going Pro: Keep Your Cache Warm

Here’s something most users don’t realize: LLM providers cache your system prompt prefix. If your context stays stable (sameAGENTS.md, same SOUL.md, same memory), every follow-up message in the session hits the cache — making it significantly cheaper and faster. This matters even more on SiliconFlow. SiliconFlow consistently delivers high cache hit rates — for example, according to OpenRouter’s provider performance data, SiliconFlow achieves the highest cache hit rate among all GLM-5.1 providers at 88.6%, significantly ahead of the competition. For developers, this translates directly to faster response times and lower inference costs on repeated contexts.
Cacheglgm5 1 Copy (2)
What this means in practice:
  • Don’t change models or rewrite your system prompt mid-session
  • Use /compress when conversations get long, it trims the token count without losing key context
  • Pick the right model upfront: use a larger model for complex reasoning and architecture decisions, switch to a lighter one for formatting and boilerplate generation
Run /usage periodically to see where your tokens are going. Small habits, real savings.

​
Want More Tips?

This covers the essentials to get you productive quickly. Hermes Agent has a rich set of best practices beyond what’s listed here — including CLI power-user shortcuts, context file patterns (AGENTS.md, SOUL.md, .cursorrules), memory management, security best practices, and more. Check out the official documentation: https://hermes-agent.nousresearch.com/docs/guides/tips

​
Choose the Right Model

Not sure which model to use? Here’s a quick reference by use case.
ModelBest forKey highlights
DeepSeek-V4-FlashFast, cost-effective coding chat / large codebase1M context · 3 reasoning modes · best value in the V4 series
DeepSeek-V4-ProComplex reasoning / large codebase1M context · #1 open-source on math, STEM & competitive coding · approaches Opus 4.6
GLM-5.1Long-horizon agentic tasks58.4 on SWE-Bench Pro · long-horizon execution · iterative self-improvement
Kimi-K2.6Long-horizon tasks / Frontend generation / multi-agentAgent swarm architecture · long-horizon coding · prompt-to-frontend generation
MiniMax M2.5Coding / agentic workflow / office workSOTA Coding Tied Claude & agentic tool use · trained across 200k+ real-world environments
Qwen3.6-27Befficient, context-aware coding experiences.Flagship-level agentic coding performance
Browse the full model library: siliconflow.com/models

​
Already Using OpenRouter?

​
Bring Your Own Key

Supports BYOK (Bring Your Own Key) on OpenRouter — your requests draw from your SiliconFlow balance, with OpenRouter waiving platform fees on your first 1M BYOK requests/month Browse the full model library → siliconflow.com/models
Byokor

​
Resources

​
Hermes Agent

​
OpenRouter

​
SiliconFlow