Hermes vs Devin comes down to where the agent lives: Hermes Agent runs on a machine you control and reaches you through chat apps and schedules, while Devin runs in Cognition's cloud on its own virtual machine and turns engineering tickets into pull requests.
So I don't set them up the same way.
Hermes gets configured like a teammate who sits in every chat.
Devin gets configured like a new developer who needs a laptop, repo access and a well-written ticket.
Neither vendor publishes a comparison, so everything below is my reading of the Hermes Agent docs and the Devin docs.
If you'd rather have my team map it onto your stack, get your Backlink Gameplan, or try Polar Rank free for 3 days.
The setup split in one table
| Setup step | Hermes Agent | Devin |
|---|---|---|
| Install | One-line installer on Linux, macOS, WSL2 or native Windows | Sign up at app.devin.ai, plus the optional Devin CLI |
| Where jobs run | Your laptop, a server, Docker, SSH or a cloud sandbox | A Linux VM booted from your saved snapshot, with Windows and macOS options |
| First thing to configure | A model provider, via hermes model | A Git connection, then the environment blueprint |
| Reusable procedures | Skills in ~/.hermes/skills/ | Skills in .agents/skills/ or on the Customize page |
| Timed work | Cron jobs delivering to a chat, a file or a platform | Automations with a Schedule trigger |
| Where you talk to it | Telegram, Slack, WhatsApp, email, Desktop, CLI and more | The web app, Slack, Teams, the CLI and Devin Desktop |
| Finished job looks like | A message, a report, a file or a completed command | A pull request, a review or a tested change |
What you're installing on each side
Hermes is an open-source agent from Nous Research, published under the MIT license in the hermes-agent repo.
The current release is v0.23.0, "The Hearth Release", shipped on 5 November 2026 according to the release notes.
You get a CLI, a terminal UI, a Desktop app, a messaging gateway, cron, skills, memory and Bot Mode in one install.
Devin is Cognition's autonomous AI software engineer, built to write, run and test code.
Its docs size jobs with a simple rule: work you could finish inside roughly three hours is usually in range.
Devin now ships as three surfaces: the cloud app, Devin CLI, and Devin Desktop, which is the name Windsurf took on 2 June 2026 per the Devin Desktop FAQ.
That's gotcha number one: any older tutorial that says Windsurf now means Devin Desktop.
Hermes setup: install, pick a model, open the gateway
Here's how I'd take Hermes from zero to useful.
- Run
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashon Linux, macOS or WSL2, or the PowerShell one-liner on Windows, both listed on the installation page. - Run
hermes modeland connect a provider: Nous Portal, OpenRouter, Anthropic, or a self-hosted endpoint such as Ollama or vLLM from the providers page. - Run
hermes gateway setupand add one chat platform, such as Telegram or Slack. - Set an allowlist so only your people can message it, for example
SLACK_ALLOWED_USERSorEMAIL_ALLOWED_USERS. - Start the gateway with
hermes gatewayand send it a test message from your phone.
Planning to use the Desktop app?
Check the floor first: v0.23.0 needs macOS 13 or later, or 64-bit Windows.
Devin setup: connect Git, then let it build its own machine
Devin needs repo access before anything else works.
- Open Settings, then Connections, and connect GitHub, GitLab, Bitbucket or Azure DevOps, choosing which repositories Devin can see.
- Start a session and type: "Set up your environment for this repo."
- Review the blueprint Devin proposes as suggestion cards and approve the ones you want.
- Let the build finish, because it produces the snapshot every later session boots from.
- Add API keys and logins as encrypted Secrets, which Devin injects as environment variables during builds and sessions.
The environment setup page calls this the single highest-leverage step for Devin, and I'd treat it that way.
Each session boots a fresh copy of that snapshot, so changes inside one session never leak into the next.
For the first real task, start in Ask mode to plan, then switch to Agent mode to execute, as the first-session guide suggests.
🔥 Want ready-made Hermes setups instead of a blank config? 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 swapping agent setups. → Join the AI Profit Boardroom
Teach both agents your procedures with SKILL.md
This is where the two tools overlap the most.
Both read skills as SKILL.md files that follow the open Agent Skills standard.
Hermes keeps them in ~/.hermes/skills/, loads them on demand, and can write new ones itself after a hard task, per its skills docs.
Devin reads them from .agents/skills/<name>/SKILL.md in any connected repo or from the Customize page, per the Devin Skills page.
Devin also proposes new skills mid-session and offers a Create PR button to commit them.
Gotcha number two: Devin's older Knowledge and Playbooks features are folding into Skills, and new accounts have started with Skills only since 24 October 2026.
Write every new procedure as a skill on both sides and leave the legacy features alone.
Scheduling: Hermes cron vs Devin Automations
Hermes cron takes plain language, like /cron add "every 2h" "Check server status", or a standard cron expression.
A job can attach skills, pin its own model, and deliver to the chat it came from, a local file, or another platform, per the cron docs.
A webhook route can also fire a cron job the moment an outside event lands.
On Devin, the old Schedules page is legacy, and new timed work is built as an Automation with a Schedule trigger.
Automations also fire on Slack messages, GitHub events, Linear and Jira changes, PagerDuty incidents and custom webhooks, per the Automations page.
Gotcha number three: sessions started by Automations don't read or write your personal Devin Memory, so anything they must know belongs in a skill.
On the Hermes side, a cron run can't create more cron jobs, which blocks runaway scheduling loops.
Wiring chat: the Hermes gateway vs tagging @Devin
The Hermes gateway is one background process connected to every platform you configure, from Telegram and WhatsApp to Microsoft Teams and email, as the messaging overview lists.
Devin's chat story is narrower and sits deeper inside engineering teams.
You install the Devin app in Slack, link your user profile once, then tag @Devin in any thread.
Keywords such as !ask for a quick codebase answer or !ultra for the most capable agent change how the session starts, per the Slack integration page.
Microsoft Teams works the same way, with an @Devin tag.
Guardrails I'd switch on first
On Hermes, I'd keep approvals.mode on smart or move it to manual, so risky shell commands need my yes, as the security page describes.
Those approvals cover dangerous commands, not outgoing messages.
So for any job that writes to customers, I put "draft only" into the job text and send the message myself.
Logins go into the Hermes passwords and logins vault, which encrypts them on the machine and keeps them away from the model.
On Devin, I'd create a dedicated account such as devin@company.com on each tool, as the Secrets page recommends, and save its logins as Secrets.
For shared MCP connections, Devin's docs strongly recommend a service account rather than anyone's personal login.
And every code change still lands as a pull request, so a human decides what merges.
Worked example: a five-person web agency
Picture a small agency that builds and maintains client websites from GitHub repos.
Here's how I'd split its week.
- A client reports a broken contact form, and the account manager files a Linear ticket.
- The ticket gets assigned to Devin, which starts a session using the default skill set in its Linear settings.
- Devin boots its snapshot, reproduces the bug in its browser, fixes it and opens a pull request.
- A developer reviews the pull request and merges it.
- On Friday afternoon, a Hermes cron job reads the week's notes folder and posts a short update per client into the team's Slack channel, written as drafts only.
- The account manager edits each update and sends it to the client.
Devin handles the code.
Hermes handles the conversation around the code.
📞 Run an agency and want this split mapped to your clients? Get your Backlink Gameplan and my team will walk through which jobs belong with which agent, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days
Worked example: a plumbing office with no code at all
Now take a two-van plumbing business with no repos and no developers.
- Install Hermes on the office PC and connect it to Telegram through the Telegram gateway guide.
- Add a skill holding the call-out checklist the office already uses: address, problem, access notes, preferred time.
- After each job, the plumber sends a voice memo, the gateway transcribes it, and Hermes files a tidy job note.
- A 7am cron job posts the day's jobs and open follow-ups from the shared notes folder to the owner on Telegram.
- Replies to customers come back as drafts only, and the owner sends them.
Devin has nothing to do here, because its whole design assumes a codebase.
That's no knock on Devin; it's a sign you'd be buying the wrong tool for this shop.
Bridging the two through the Devin MCP server
You don't have to pick one if you run both.
Devin publishes an official MCP server at https://mcp.devin.ai/mcp that lets MCP-compatible agents manage Devin sessions and schedules, per the Devin MCP page.
Hermes connects to remote HTTP MCP servers with a URL and headers in config.yaml, per its MCP docs.
On paper, that lets a Hermes Bot open a Devin session from a Slack or Telegram request.
Neither vendor documents this pairing as a recipe, so treat it as two documented parts bolted together.
mcp_servers:
devin:
url: "https://mcp.devin.ai/mcp"
headers:
Authorization: "Bearer ${DEVIN_API_KEY}"
trust: untrusted
Use a Devin service user key, keep it in an environment variable, and leave trust: untrusted on so Hermes asks before every write call, as the MCP config reference explains.
Gotchas I'd write on a sticky note
- Windsurf is Devin Desktop now, and your Windsurf settings and rules carry over.
- Knowledge is deprecated and Playbooks are becoming Skills, so build skills.
- There's no Devin 3: the last numbered release was Devin 2.2 on 24 February 2026, per the 2026 release notes.
- Hermes webhook payloads can carry hostile instructions, so the docs say to run an internet-facing gateway in Docker or over SSH.
- Hermes v0.23.0 removed the old
browser_*tools, so prompts that name them should switch tobrowser_exec. - Devin's Computer Use toggle is admin-only and sits under Settings, Devin, Sessions.
Verdict by job
Pick Devin if the job starts as a ticket in Linear, Jira or GitHub and should end as a reviewed pull request.
Pick Devin if each task needs a fresh, reproducible machine with a desktop it can click through for testing.
Pick Hermes if the job starts as a chat message, an email or a schedule and ends as a reply, a report or a file.
Pick Hermes if you want to choose the model, host it yourself, or run on local hardware.
Run both if you have code and customers, and let the Devin MCP server be the handoff point.
Want the Agent OS version of this setup? Join the AI Profit Boardroom and grab the walkthroughs.
For the code-editor angle on the same question, read my Hermes vs Cursor setup guide.
FAQ: Hermes vs Devin
Can Hermes start a Devin session?
Devin's official MCP server lets any MCP-compatible agent manage sessions, and Hermes connects to remote HTTP MCP servers. Neither vendor documents the pairing as a recipe, so set the server to trust: untrusted and approve each write.
Where do skills go for each agent?
Hermes keeps skills in ~/.hermes/skills/. Devin reads them from .agents/skills/<name>/SKILL.md in a connected repo or from its Customize page.
Do I still set up Playbooks in Devin?
No. Devin is moving Playbooks to Skills and deprecating Knowledge, and new accounts have started with Skills only since 24 October 2026.
Does Hermes ask before sending a customer message?
Its approvals cover dangerous commands, not messages. Write "draft only" into any job that touches customers and send the message yourself.
What happened to Windsurf?
It became Devin Desktop on 2 June 2026. Cognition says it is the same IDE with the Agent Command Center as the main view, and Windsurf settings carry over.
📺 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, connect one repo to Devin tomorrow, and get your Backlink Gameplan if you want my team to check your setup for hermes vs devin, or try Polar Rank free for 3 days.











