Hermes vs Perplexity Computer is a choice between an agent you install and run yourself and an agent Perplexity runs for you in its own cloud.
I'd set up both, because they solve different halves of the same week.
Hermes Agent is open-source software from Nous Research under the MIT license, and it runs on your laptop, a server, Docker or a remote sandbox.
Perplexity Computer is a hosted worker that normally runs every task in a secure cloud sandbox, with no local install needed.
Below is the order I'd set them up in, the one documented bridge between them, and the settings I'd change before either touches real work.
If you'd rather have my team map your jobs to the right tool, get your Backlink Gameplan, or try Polar Rank free for 3 days.
The one-line difference before you install anything
With Hermes, you own the box.
You pick the model, the machine, the chat apps and the approval rules, and everything sits in config files under ~/.hermes/.
With Perplexity Computer, Perplexity owns the box.
You describe an outcome, Computer splits it into subtasks, and sub-agents work through them in an isolated environment that has a filesystem, a browser and your connected tools, as the launch post explains.
That single split, self-hosted versus hosted, drives almost every setting further down.
Hermes vs Perplexity Computer side by side
| Area | Hermes Agent | Perplexity Computer |
|---|---|---|
| What it is | Open-source, self-improving agent (MIT) | Hosted digital worker inside Perplexity |
| Where tasks run | Local, Docker, SSH, Modal, Daytona or Vercel Sandbox backends | Cloud sandbox by default; Personal Computer and Portable Computer for local work |
| Models | Any provider you set with hermes model | Chosen for you by effort setting, with a Custom mode |
| Where you talk to it | CLI, TUI, Desktop, Webapp, 20+ chat platforms | Web, iOS, Android, desktop app, Slack, email |
| Recurring work | Cron in plain language or cron syntax, plus webhook-fired jobs | Automations, as often as every 15 minutes, plus app triggers |
| Extra tools | MCP client (stdio and HTTP), skills, plugins | Connectors, skills, Plugins, custom remote MCP connectors |
| Access | Free software; you bring model access | Paid Perplexity plans (Pro, Max and Enterprise tiers) |
Neither Nous Research nor Perplexity publishes a comparison of the two, so this table is my own reading of each vendor's docs.
Step 1: install Hermes and choose a model
On Linux, macOS or WSL2, one line installs it, and a second picks your model provider.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes model
Native Windows uses a PowerShell one-liner instead, and the installation page also covers the Desktop app, Docker, Nix and Termux.
If you want the Desktop app, the v0.23.0 notes say it now needs macOS 13 or later, or 64-bit Windows.
The providers page lists Nous Portal, Anthropic, OpenAI, OpenRouter, xAI, Google Gemini, Kimi, MiniMax, local Ollama or LM Studio, and any custom endpoint.
Switching later is one command, so don't overthink the first pick.
Step 2: start your first Perplexity Computer task
There's nothing to install for the cloud version.
Open the Home page, select the Computer icon, and type into the Start a task field.
Once a task starts in Computer, that session can't be switched back to Search, so start a fresh session from Home if you only wanted an answer.
The product page says Computer is available to Pro and Max subscribers, and some free accounts get a small one-time allowance of trial turns.
Put everything Computer needs into your first message on a trial account, because every follow-up uses a turn.
🔥 Want my exact Hermes config and the prompts I hand to hosted agents? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members comparing setups like this one. → Join the AI Profit Boardroom
Step 3: give each tool a place to receive work
Hermes reaches people through its messaging gateway, one background process that connects every platform you configure.
hermes gateway setup
hermes gateway install
The gateway docs cover Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, Microsoft Teams, Matrix, LINE and more.
Lock it down with an allowlist such as TELEGRAM_ALLOWED_USERS so only your own accounts can give it orders.
Perplexity Computer comes to you through its own surfaces instead.
A Slack admin installs it once from the Slack Marketplace, then people mention it in a thread, and Perplexity's Slack guide says it reads the earlier messages in that thread.
For email, you send a task to computer@perplexity.com from the address tied to your account, and Computer in Email replies in the same thread with attachments.
The address ends in .com, and the help page says the .ai version is not supported.
Step 4: schedule the recurring jobs
In Hermes, you can schedule from chat, from the CLI or just by asking.
/cron add "every 2h" "Check the shared inbox and summarize anything urgent"
hermes cron create "every 1h" "Summarize new feed items" --skill blogwatcher
The cron docs say a job can attach skills, deliver to the chat it came from or any configured platform, and fire from a webhook route the moment something happens.
There's also a no-agent mode that runs a plain script on a schedule with no model involved.
Perplexity's version is called Automations, and you create one by describing the task and cadence in plain language.
The Automations guide lists schedules as often as every 15 minutes, plus conditional triggers for new Gmail or Outlook mail, Slack messages, Google Calendar events, Google Drive files and Linear issues.
Each run uses credits, and Computer reminds you of that before it saves the schedule.
I go deeper on scheduling on the Hermes side in my Hermes cron job walkthrough.
Step 5: add your own tools over MCP
Hermes reads MCP servers from ~/.hermes/config.yaml, and local stdio servers sit next to remote HTTP ones.
mcp_servers:
filesystem:
command: "npx"
args: ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/projects"]
The Hermes MCP page also lets you filter each server so the agent only sees the tools you want it to.
Perplexity Computer takes remote MCP servers as custom connectors.
Open Plugins, choose Add custom, then Add MCP connector, and the connector guide asks for a name and an https URL, with OAuth, API key or no auth, over Streamable HTTP or SSE.
Inside an organization, an admin has to switch on member-added connectors first.
📞 Not sure which jobs belong on your own server and which belong in Perplexity's cloud? Get your Backlink Gameplan and my team will look at your workflows with you, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days
Step 6: wire Perplexity's API into Hermes
This is the only link between the two that either vendor documents.
The Hermes providers cookbook includes Perplexity as a custom provider, which it describes as useful when you want a model that searches the live web and cites sources.
# ~/.hermes/config.yaml
providers:
perplexity:
api: https://api.perplexity.ai
key_env: PERPLEXITY_API_KEY
model:
default: sonar
provider: custom:perplexity
The key goes in ~/.hermes/.env as PERPLEXITY_API_KEY.
Be clear about what this is: Perplexity's API models running inside Hermes, not Perplexity Computer itself.
I'd keep it as a second provider and switch to it per turn with /model custom:perplexity:sonar for research questions.
Approval settings I'd change on day one
Hermes checks every shell command against a list of dangerous patterns before running it.
Its security page offers three modes for approvals.mode: smart (the default), manual and off.
I'd set manual while you're learning, and leave cron_mode: deny as it ships so a scheduled job can't approve its own risky command.
Avoid YOLO mode outside a throwaway container.
v0.23.0 also added price tags and spend limits for paid MCP tools, and the release notes say Hermes asks before going over a chat's limit even in YOLO mode.
On the Perplexity side, Personal Computer uses on-device authorization for sensitive steps, and remote tasks still follow the device's own approval rules.
Where your files actually live in each setup
Hermes works wherever its terminal backend points, so a local backend means local files and a Docker backend means a container.
Perplexity has two local options on top of the cloud.
Personal Computer is the Mac and Windows desktop app, on macOS 15 or later and Windows 10 or 11, with local folders, voice and native Mac app control, which isn't on Windows yet.
Portable Computer runs the model and agent on an NVIDIA DGX Spark or a supported RTX PC, and asks before sending any step to the cloud.
If client files must never leave your hardware, both Hermes with a local model and Portable Computer can do that, but Hermes does it on more kinds of machine.
Operator example: a plumbing firm's morning job sheet
Here's how I'd split one real routine for a small plumbing business.
Hermes runs on the office PC with the gateway connected to the owner's Telegram.
Step one, a 7am cron job reads the bookings folder and writes the day's job list.
Step two, it groups jobs by area and drafts a text for each customer saying when the plumber will arrive.
Step three, it sends the draft list to the owner on Telegram, and nothing goes to customers until the owner says yes.
Perplexity Computer takes the research side.
Step one, the owner asks Computer to compare three suppliers for a boiler part and build a short table.
Step two, Computer runs the searches in parallel and returns a cited sheet.
Step three, the owner saves it as an Automation that re-checks the same suppliers every Monday.
Agency example: weekly client research packs
A small marketing agency can run the same split.
Computer gets a Monday Automation that researches each client's competitors and emails a cited summary.
Hermes gets the client-facing jobs: it watches the agency's Slack, drafts status updates from the week's notes and keeps each client's preferences in its memory files.
The account manager approves anything before it reaches a client.
Gotchas that catch people out
- The Hermes Webapp is off by default; turn on
webapp.enabledor Web Access before you expect a browser tab. - The old built-in browser tools were removed in v0.23.0, so prompts that name
browser_navigateneed switching tobrowser_exec. hermes mcp serveis nowhermes mcp-serve, and nothing rewrites existing client configs for you.- Perplexity's effort selector lives on web, iOS and Android; the Mac and Windows apps still show the regular model picker.
- If a Perplexity Automation keeps skipping runs, it pauses and stays paused until you press Enable.
My verdict by job type
Pick Hermes when you want control: your model, your server, your chat apps, approvals you set, and an agent that stays running on hardware you manage.
Pick Perplexity Computer when you want research-heavy work done with zero setup, cited answers, and connectors you sign into with a click.
Pick both when you want Hermes as the always-on operator and Computer as the research desk, linked through nothing more than you reading Computer's output and the Perplexity API sitting inside Hermes.
The configs, prompts and walkthroughs I use for this are inside the AI Profit Boardroom.
FAQ: Hermes vs Perplexity Computer
Can Hermes control Perplexity Computer directly?
Neither vendor documents a direct link. The documented bridge is the Hermes providers cookbook, which adds Perplexity's API models (such as sonar) as a custom provider inside Hermes.
Which file holds the Perplexity API key for Hermes?
Put PERPLEXITY_API_KEY in ~/.hermes/.env and reference it with key_env in the providers block of ~/.hermes/config.yaml.
What is the shortest schedule each tool allows?
Perplexity Automations run as often as every 15 minutes. Hermes cron takes natural-language intervals or cron expressions, and can also fire instantly from a webhook route.
Which approval mode should a new Hermes install use?
Start with approvals.mode set to manual so every dangerous command asks you first, and keep cron_mode on deny so scheduled jobs can't approve themselves.
Does Perplexity Computer need anything installed?
Not for the cloud version, which runs from the web, iOS or Android. Personal Computer and Portable Computer are the installable options for local files and on-device models.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Install Hermes tonight, start one Computer task tomorrow, and get your Backlink Gameplan if you want help settling hermes vs perplexity computer for your own jobs, or try Polar Rank free for 3 days.











