Five tools that make Hermes worth using every day. RTK saves money, Hindsight adds memory, Tokscale shows what you spend, Ghostty upgrades your terminal, and Camoufox gives your agent a browser.
—
You ran hermes in the terminal and got a response. Congratulations, you are in the door. But running an AI agent and getting real value from one are two different things. The gap is not the model. It is the tooling around it.
After months of using Hermes daily, I have narrowed my setup to five tools that actually changed how I work. Four of them make the experience noticeably better. One saves real money. Here they are in the order I would recommend installing them.
Ghostty , Upgrade Your Terminal First
If you spend most of your time in a terminal running Hermes rather than inside a desktop IDE plugin, the first thing worth swapping is not the agent. It is the terminal itself.
Ghostty is what I kept after trying iTerm2, Warp, Kitty, and Alacritty. The three things that won me over: speed, simplicity, and native feel. It is written in Zig with GPU acceleration, so startup time is in a different league from iTerm2. Configuration is straightforward too. Edit ~/.config/ghostty/config directly and you are done. No GUI settings to dig through.
font-family = JetBrains Mono
font-size = 15
theme = catppuccin-mocha
window-padding-x = 8
window-padding-y = 8
Warp tries to hide complexity from you. iTerm2 often has more features than you need with configuration scattered everywhere. Ghostty splits the difference. It leaves you in control while handling performance and feel. If you live in the CLI with Hermes, this few minutes is worth it.
Tokscale , See What You Are Actually Spending on Tokens
Once Hermes becomes part of your daily workflow, the question hits fast. How much did I actually spend on tokens this week? Most people cannot answer that. Not because they are not tracking it. The data is just scattered across ~/.hermes/, ~/.claude/projects/, and whatever else you have open simultaneously.
Tokscale solves this directly. It scans local data from over twenty AI coding agents and pulls everything into a single terminal dashboard. No installation required, one command gets you started.
npx tokscale@latest
Requires only Node.js, Bun, or Deno. It automatically finds data from Hermes, Claude Code, Codex, Cursor, and whatever else is on your machine, then opens an interactive TUI. The first screen gives you total tokens, a cost estimate, and a breakdown of which models are burning through your budget fastest. The problem is rarely that you are not spending money. It is that you cannot see it in one place.
# Hermes usage only
tokscale -client hermes
# Export to JSON for your own dashboard
tokscale models -month -json > report.json
The GitHub-style contribution heatmap is the detail I keep coming back to. Daily AI coding activity shown as colored blocks, available in both 2D and 3D. That vague feeling of this week was busy suddenly becomes concrete when you see it as a visual grid.
Two caveats worth knowing. First, costs are API-rate estimates. If you are on a flat subscription like Claude Max or Cursor Pro, Tokscale will read higher than your actual monthly bill. It is best used for comparing relative spend across tools, not as a billing tracker. Second, it runs fully local by default, no login required for analysis. Only the leaderboard and Wrapped sharing features need a GitHub account.
Hindsight , Replace the Built-in Memory System
Hermes ships with a built-in MEMORY.md file. It works, but it is thin. It only writes when it decides something is important, has a hard character cap around 2,200 tokens, and does not proactively extract entities or relationships from your conversations. For ongoing collaboration, memory just is not enough.
Hindsight fills that gap. After each conversation, it automatically extracts entities, facts, and relationships with timestamps. Before your next session, it injects relevant memories back into context.
Setup is simple.
hermes memory setup
Select hindsight from the wizard, it handles the dependency installation for you. Then register at https://ui.hindsight.vectorize.io/connect to get an API key. The free tier is sufficient for personal use.
Verify it is working.
hermes memory status
When you see Hindsight active with bank_id, auto-recall, and auto-retain status fields populated, you are done. The immediate difference after switching: Hermes starts remembering who you are. You stop repeating the same context-setting introductions at the start of every session.
Camoufox , Give Your Agent a Real Browser
For an agent to actually work on the web, reading page text is not enough. It needs to see pages, click buttons, fill forms. Camoufox is a browser built specifically for AI agents. It is based on Firefox with modifications at the browser level: fingerprint spoofing, decluttering, and performance tuning. The practical upside is lower detection rates on sites that try to identify automation, and memory usage that stays well below what most people expect.
Install it.
pip install -U "camoufox[geoip]"
python -m camoufox fetch
Usage.
from camoufox.sync_api import Camoufox
with Camoufox() as browser:
page = browser.new_page()
page.goto("https://example.com")
print(page.title())
It is Playwright-compatible, so if you already have scripts written for Playwright, migration is minimal. One boundary to keep in mind: Camoufox is not a universal bypass tool. Detection systems evolve. It handles personal agent use and lightweight automation fine. For production-grade work, you still need your own rate controls and compliance checks.
RTK , This One Saves Money Directly
If you are going to install only one tool from this list, make it RTK.
RTK (Rust Token Killer) does one thing. Terminal command output is frequently hundreds of lines, and a large portion of that is not what you actually care about. File listings from ls, untracked files from git status, detailed logs from cargo test. RTK filters the low-value output before it reaches your context window.
The impact is real. It cuts token consumption from terminal tools by 60 to 90 percent. For an agent that runs shell commands frequently, that is direct dollar savings.
Install.
brew install rtk
Then initialize.
rtk init -g
This installs a global hook. From that point on, whenever Hermes calls a shell tool, the command output passes through RTK automatically.
Verify.
rtk -version
rtk gain
The rtk gain command shows how many tokens you have saved. The first time you run it, you will probably wish you had installed it sooner.
One clarification that is easy to miss. RTK only compresses terminal tool output. It does not change how Hermes communicates with the LLM. It reduces the context cost of doing the same work, not by doing less work. In developer scenarios, those gains are clean, almost no downside.
Quick Summary
If you want the shortest possible path: install RTK first and stop there if you have to. If you want to actually get comfortable with Hermes daily, the full stack in priority order is: RTK for the budget, Hindsight for memory, Tokscale for visibility, Ghostty for the terminal feel, Camoufox for browser capability.
None of these are plugins or features bolted onto an existing product. They are tools that fix specific problems with how AI agents actually get used in practice. The five that made the cut are the ones I have kept running for months without reaching for alternatives.