⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB

GoldieBench Blog · 7 min read

Hermes Agent Microsoft Teams Setup: Register The Bot, Connect The Meeting Pipeline, Then Pick A Model With Real GoldieBench Scores

Stage one puts Hermes in your Teams chats. Stage two turns meetings into summaries. A short last section covers which model to run behind it.

Illustrated hero: a dark navy panel showing a gold chat bubble connected by a glowing line to a meeting transcript card, with a small leaderboard of gold bars

You set up hermes agent microsoft teams in two stages: first you register Hermes as a Teams bot and point it at your gateway, then, if you want meeting write-ups, you enable the teams_pipeline plugin and connect it to Microsoft Graph.

This guide covers both stages in order, from the Hermes Teams setup docs and the Teams Meetings docs, with one short section at the end on which model to run behind it.

The running example is a ten-person recruitment firm that already holds every candidate call in Teams.

If you'd like my team to plan it with you, book a free SEO strategy session.

Before You Start: What You Need On Hand

  • A working Hermes install with the gateway on a machine that stays online.
  • Node and npm, for Microsoft's Teams CLI.
  • A way to publish one HTTPS address: a tunnel for testing, a domain with a valid certificate later.
  • The AAD object IDs of everyone who should use the bot.
  • For meeting summaries only: a tenant admin and Teams licences that produce transcripts and recordings.

The bot stage needs none of the admin items, so the recruitment firm can start it today.

Stage 1, Step 1: Register The Bot

# one-time install of Microsoft's CLI
npm install -g @microsoft/teams.cli@preview
# sign in with your work account
teams login
# shows your AAD object ID for the allowlist
teams status --verbose
# register the bot against your public URL
teams app create --name "Hermes" --endpoint "https://<public-url>/api/messages"

The CLI takes care of registration, so there are no Azure portal screens.

Its output holds three credentials (client ID, secret, tenant ID) plus the link you'll use to install the app.

Copy the secret at once; the CLI won't display it again.

Stage 1, Step 2: Make The Gateway Reachable

Teams calls your bot over the internet, so a localhost address won't do.

Point a tunnel at port 3978, the adapter's default, and use its HTTPS URL as the endpoint above.

The tunnel handles TLS; the Hermes listener behind it speaks plain HTTP.

For the long run, swap the tunnel for a reverse proxy on your own domain and run teams app update with the new endpoint, since Teams refuses self-signed certificates.

Stage 1, Step 3: Configure, Start And Install

# ~/.hermes/.env
TEAMS_CLIENT_ID=<client-id>
TEAMS_CLIENT_SECRET=<client-secret>
TEAMS_TENANT_ID=<tenant-id>
TEAMS_ALLOWED_USERS=<id-1>,<id-2>
# restart, then confirm the listener answers ok
hermes gateway restart
curl http://localhost:3978/health

# print the link that installs the bot in Teams
teams app get <teamsAppId> --install-link

A healthy listener answers ok.

Open the install link, which lands in the Teams client, and send the bot a private message.

The firm's recruiters then add it to their Candidates channel and @mention it there, because in channels and group chats it stays quiet unless tagged.

Prefer a wizard? hermes gateway setup with Microsoft Teams selected walks the same path.

Stage 1, Step 4: Settle The House Rules

Keep the allowlist filled; the docs warn an empty one lets anyone who finds the bot use it.

Risky commands appear as an Adaptive Card with four choices, so nothing dangerous runs on a casual reply.

By default, the security docs let any admitted member answer a pending approval, and approvals.approvers can tie each card to the requester or to named sign-off users.

Know the limits too: per the platform table, Teams chats handle pictures, threaded replies and the typing indicator, while voice notes, file uploads, reactions and streamed answers are missing.

Lock the env file with chmod 600 ~/.hermes/.env.

When The Bot Doesn't Answer: Quick Fixes

  • Silence from one person only: their AAD object ID isn't in the allowlist yet. Add it and restart.
  • Silence in a channel: nobody tagged the bot. It needs an @mention outside private chats.
  • A 400 line in the log: HTTPS is reaching the plain HTTP listener. End TLS at the tunnel or proxy.
  • "This bot is not responding" in Teams: the webhook threw an error; hermes gateway status -l shows the traceback.
  • Auth failures: recheck the three TEAMS_ values and that the tenant matches the account used for teams login.
  • Everything broke after a restart: your tunnel may have handed out a new URL; a named devtunnel keeps the same one.

The health check on port 3978 is the fastest first test whenever something looks off.

If it answers ok but Teams still gets nothing, the problem is almost always between Teams and your public address.

Stage 2: Turn Meetings Into Summaries

With the bot stable, the firm's admin builds the Graph side from the Graph app registration guide:

  1. Register a single-tenant app and create a client secret.
  2. Grant OnlineMeetings.Read.All and OnlineMeetingTranscript.Read.All, plus the recording permissions for a fallback, then click Grant admin consent.
  3. Limit the app to the recruiters' meetings with an Application Access Policy.
  4. Run hermes plugins enable teams_pipeline and add the MSGRAPH_ credentials.
  5. Enable the msgraph_webhook listener on port 8646 with a strong client state value.
  6. Pick graph or incoming_webhook delivery and subscribe to transcripts with hermes teams-pipeline subscribe.

After each call, Graph pings the listener, the pipeline grabs the transcript, and the summary posts to the channel you chose, or to Notion or Linear.

No transcript? It downloads the recording and runs speech-to-text instead, provided ffmpeg is installed.

For the recruitment firm, graph delivery makes sense, because summaries can land as threaded posts under the app's name in the Candidates channel.

Incoming webhook delivery is simpler to start with: the channel owner creates a URL, and posts show under the webhook's identity without reply threading.

Run hermes teams-pipeline validate after each change, and token-health --force-refresh if Graph sign-in looks stale.

Stage 2, Last Step: Keep Subscriptions Alive

Graph subscriptions end after 72 hours at most, and none of them renew on their own.

The operator runbook offers Hermes cron, a systemd timer or crontab to run hermes teams-pipeline maintain-subscriptions, and suggests a 12-hour cycle.

Check it with hermes teams-pipeline subscriptions and the --dry-run flag, and look at hermes teams-pipeline list --status failed each week.

Microsoft adds one more condition: an admin can disable Graph access to transcripts, which makes transcript subscriptions fail.

Adding More Recruiters Later

Growing the bot to the whole firm is a config change, not a rebuild.

Append each new recruiter's AAD object ID to TEAMS_ALLOWED_USERS, comma-separated, and restart the gateway.

For meeting summaries, grant the same Application Access Policy to each new recruiter with Grant-CsApplicationAccessPolicy, then allow up to 30 minutes before testing.

Set TEAMS_HOME_CHANNEL to the Candidates channel if you also want scheduled Hermes jobs to post there.

Review the list whenever someone leaves, so access always matches the team.

Which Model To Run Behind It

The setup above doesn't depend on a model, but the quality of chat replies and summaries does.

On the live GoldieBench leaderboard read on 15 November 2026, the top single models were Claude Opus 5 at 8.27 and GPT-5.6 Sol at 8.16, with Fusion leading overall at 8.59 as a combined OpenRouter entry.

The methodology scores one-shot HTML builds, not summaries, so use it to shortlist and then compare two models on three real transcripts.

My Verdict

Do stage one this week: CLI, tunnel, env file, install link, allowlist.

Do stage two once your admin, licences and renewal schedule are ready.

For a hosted agent that joins calls, see our Grok Bot Microsoft Teams page; for help with Hermes, ask inside the AI Profit Boardroom.

Also On Our Network

Register the bot, open the endpoint, fill the env file, then add the Graph app and the renewal job, and if you want help, ask my team on a free strategy call or inside the AI Profit Boardroom to finish your hermes agent microsoft teams setup.

FAQ

What are the two stages of a Hermes Teams setup?

Stage one registers Hermes as a Teams bot through the Teams CLI and the gateway. Stage two enables the teams_pipeline plugin and connects it to Microsoft Graph for meeting summaries.

Which port does the Hermes Teams bot listen on?

Port 3978 by default, at /api/messages. Change it with TEAMS_PORT. The Graph listener for meetings uses port 8646.

Does the meeting stage need an admin?

Yes. A tenant admin has to grant admin consent for the Graph application permissions and ideally add an Application Access Policy.

What happens if no transcript exists for a meeting?

The pipeline downloads the recording and runs speech-to-text instead, as long as ffmpeg is installed.

Does the setup depend on which AI model I use?

No. The model only affects reply and summary quality. Shortlist from GoldieBench, then test on your own transcripts.

How do I stop summaries from quietly stopping?

Schedule hermes teams-pipeline maintain-subscriptions every 12 hours, because Graph subscriptions last 72 hours at most.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
4coaching calls a week