Docs in one page

Parlant Agents is a service for running managed, always-on AI agents without setting up or maintaining a server yourself.

Each agent gets dedicated infrastructure, persistent storage, a secure web address, encrypted backups, and dashboard controls. Parlant handles provisioning, health monitoring, recovery, and tested upgrades. You keep control of the agent itself: choose how it accesses AI models, configure its tools and integrations, and decide what work it may do.

Hermes by Nous Research is the first supported agent. It can chat, use tools, work with files, browse the web, remember selected context, and carry out multi-step tasks while your own computer is offline. More agent runtimes can be added after they meet the same management and recovery standards.

Parlant manages the infrastructure around your agent; it does not remove your responsibility to review important outputs, protect credentials, or approve consequential actions. Model and integration providers receive the data needed to perform requests you send to them.

Start with Getting started with Parlant Agents, or explore the feature articles to see what every managed agent includes.

This page records changes that affect customers directly. Internal maintenance and administrative changes are intentionally omitted.

July 22, 2026

• Rebuilt Guides as a structured Docs library with multi-page and one-page reading modes.
• Added localized English and German public URLs with consistent language switching.
• Added the Parlant Agents Blog and made docs and blog content easier to discover.

July 21, 2026

• Added a transparent pricing page covering bring-your-own-key, managed model credit, and multiple-agent options.
• Made the latest confirmed price available even during a temporary pricing-provider interruption.

July 16, 2026

• Added an embedded, guided checkout that shows billing details, taxes, and totals before payment.
• Published no-code personal-assistant and weekly-planning guides.
• Added replayable VPS customization scripts, with explicit controls during upgrades and recreation.

July 15, 2026

• Renamed the service to Parlant Agents.
• Added the first English and German guide library.
• Added dashboard-driven Hermes upgrades with a verified backup before server replacement.
• Expanded backup coverage to conversations, configuration, skills, work files, and integration state; added separate retention and download controls for manual and automatic backups.
• Added managed Hermes WebUI and Hermex access for browsers and iPhone.
• Added agent health status and a manual restart control.
• Enabled the persistent Hermes messaging gateway and LLM Wiki.

July 14, 2026

• Added managed OpenRouter model credit, visible balances, and optional top-ups.
• Added a setup choice between managed model credit and your own supported provider key.
• Reorganized the dashboard into focused panels with masked, copyable credentials.

July 13, 2026

• Launched the managed Hermes service with guided account setup and subscription checkout.
• Added dedicated agent infrastructure, secure HTTPS access, persistent storage, and encrypted backup and restore controls.

You can always get help by emailing [email protected].

To help us understand the issue quickly, include:

• the email address used for your Parlant Agents account;
• the affected agent hostname, if one exists;
• what you expected and what happened instead;
• the approximate time of the problem, including your time zone; and
• a screenshot or exact error message when useful.

Never email passwords, API keys, backup decryption keys, payment-card details, or other secrets. Support will ask for additional non-secret details if needed.

For setup questions, start with Getting started with Parlant Agents. For an agent that reports an unhealthy state, include that status and any restart result in your message.

Overview

Every managed agent includes hermes-webui, a browser interface for talking to Hermes and reviewing conversations. Parlant installs it, gives it a secure HTTPS address, and keeps its version compatible with the managed Hermes release.

Open the web UI

Open your agent in the Parlant Agents dashboard. Select Open agent, or copy the Hermes WebUI URL and Hermex password from the credentials panel. The same address works on phone, tablet, and desktop; no port forwarding, DNS setup, or local installation is required.

On iPhone, you can also connect the Hermex app to the managed web UI. See Hermes on iPhone and Android for the current options and setup steps.

What to know

The web UI is separate from the Parlant Agents dashboard. Use Parlant Agents to manage infrastructure, backups, upgrades, billing, and service credentials. Use Hermes WebUI to chat and manage Hermes behavior.

Treat the web UI address and password as private. Do not share screenshots that expose credentials or sensitive conversation content.

Overview

Parlant Agents pre-installs a browser so Hermes can open pages, follow links, read web content, and interact with supported sites through its tools. The browser runs on the always-on agent server, not on your laptop.

This makes source-backed research, information gathering, and repeatable web workflows available without installing a browser or automation stack yourself.

Good first tasks

Start with bounded, reviewable work:

• compare information from a short list of sources;
• collect links and quotations for a research brief;
• inspect public documentation and summarize relevant sections; or
• test a web workflow without submitting its final action.

Ask Hermes to cite URLs and separate evidence from conclusions. See Use case: a verifiable research brief for a complete workflow.

Keep consequential actions controlled

A browser can reach external services, but access still depends on site behavior, authentication, and the tools enabled in Hermes. Do not assume Hermes can see an account you have not connected. Keep purchases, publication, account changes, and messages behind explicit approval, and avoid entering secrets into untrusted pages.

Overview

Each Parlant agent runs on dedicated infrastructure and includes a skill for administering its virtual private server (VPS). You can ask Hermes to inspect the system, install or configure software, and adapt the server when a task needs more than the standard image.

This is full system access, not a simulated settings panel. Review proposed commands before approval, especially when they change networking, authentication, storage, or running services.

Make changes survive replacement

One-off system changes may be lost when the server is recreated or upgraded. Put repeatable setup scripts in $HERMES_HOME/vps-customizations/. During upgrade or recreation, the Parlant Agents dashboard lets you choose whether registered customization scripts should run on the replacement server.

A server becomes marked as customized after Hermes uses administrator access. The dashboard shows that state and requires confirmation before replacement. Old scripts can conflict with a newer managed image, so keep them small, readable, and safe to run again.

Safer workflow

1. Ask Hermes for a plan and the exact commands.
2. Create a manual backup before a significant change.
3. Test one small change and verify the result.
4. Save only repeatable setup as a customization script.
5. Keep credentials outside scripts and versioned files.

Overview

Parlant Agents provides a complete Hermes installation rather than a reduced chat interface. You can use Hermes configuration, tools, plugins, memory, skills, scheduled tasks, and messaging channels supported by the installed release.

Parlant manages the server around Hermes but does not lock its application settings. You decide which capabilities to enable and which external services to connect.

Two dashboards, two jobs

Use the Parlant Agents dashboard for infrastructure concerns: service status, credentials, managed model credit, backups, upgrades, restart, and subscription management.

Use the Hermes dashboard and WebUI for agent concerns: model selection, provider keys, tools, plugins, memory, skills, and messaging configuration.

You can choose managed OpenRouter credit during setup or bring your own supported provider key. See Model access, credits, and billing before connecting a provider.

Configure deliberately

Enable one integration at a time and test it with a low-risk task. External model, messaging, and tool providers receive the data required for requests sent to them. Give each integration the least access it needs, never store secrets in agent memory, and keep consequential actions behind explicit approval.

Introduction

Your agent accumulates conversations, configuration, skills, working files, provider keys, channel credentials, and tool configuration. Parlant Agents creates encrypted backups so that infrastructure maintenance or cancellation does not have to mean losing that state.

Examples

• Before a major configuration change, create a manual backup from the agent dashboard.
• During a managed Hermes upgrade, Parlant Agents verifies a fresh backup before replacing the server.
• When you cancel and destroy an instance through the normal path, the service creates and verifies a final backup before teardown.

In-depth guide

Automatic and manual backups

Automatic backup runs daily at midnight UTC. To request an additional backup, open your agent in the Parlant Agents dashboard, select Backups, and then select Back up now. The status changes while the backup is created, uploaded, and verified.

A backup is ready for recovery only after verification completes. The service independently retains the latest manual backup and latest automatic backup for each instance. Automatic includes scheduled, pre-upgrade, pre-recreation, and pre-destruction backups. Completing a newer backup replaces only the older backup of the same kind.

The dashboard shows each retained backup's completion time, encrypted size, and SHA-256 digest.

Backups include the Hermes home, conversations and databases, profiles, skills, channel configuration, files workspace, API keys entered in Hermes, and user-home state used by Hermes integrations such as SSH, AWS, Claude, Codex, and other tool credentials. Regenerable caches and logs are omitted. Machine identity and Parlant-managed tunnel, dashboard, WebDAV, and backup credentials are regenerated securely during restoration rather than copied from the old server.

Managed upgrades

An upgrade follows a guarded sequence: create and verify a backup, remove the old server, create a server from the current managed image, and restore your data. The hostname, tunnel, credentials, and subscription remain attached to the instance.

Hermes has administrator access to its dedicated server. Using that access marks the instance as customized. Before upgrading a customized instance, the dashboard warns you and requires explicit confirmation. The customization switch defaults on for a customized instance. Turn it off for a clean system boot: scripts remain in $HERMES_HOME/vps-customizations/, but are not run as root and the successfully recreated server is no longer marked as customized. One-off commands are never replayed, and old scripts may conflict with a newer image.

The Recreate tab performs the same backup-first server replacement using the instance's existing managed image version instead of the newest image. A customized instance gets the same replay switch and confirmation there.

Do not begin an upgrade or recreation while important agent work is still running. Wait until the dashboard reports that Hermes is ready before resuming work.

Cancellation and destruction

Manage renewal through the Stripe billing portal. Cancellation takes effect according to the billing period shown there. Instance destruction is a separate, explicit action in the Parlant Agents dashboard.

The normal destruction path creates a final backup first. The emergency Destroy without backup path permanently removes the current volume without making that recovery copy; use it only when data must be removed immediately.

After a subscription expires without renewal, retained backups remain available until seven days after that subscription period ends. During this grace period you can download a backup or start a replacement subscription and restore it. After the grace period, backup objects and metadata are automatically removed. Renewing or replacing the subscription before cleanup keeps backups attached to the active instance.

Restoring an agent

After teardown, open Backups in your account. If you still have an unused active subscription, restoration can use it without purchasing another one. Otherwise, the interface offers a new subscription before restoration. Follow the live provisioning status until the recovered Hermes dashboard is ready.

Downloading and opening a backup

Open Backups, choose the manual or automatic backup, and select Download .tar. Copy the decryption key shown for that instance and store it separately from the downloaded file. Anyone who has both can read the backup.

First compare the downloaded file's digest with the SHA-256 value shown in the control plane:

sha256sum backup.tar

On macOS, use shasum -a 256 backup.tar instead. Do not continue if the digest differs.

On a Linux or macOS system with Bash, OpenSSL, jq, and tar with bzip2 support, extract and verify the encrypted payloads:

mkdir -p backup-work restored

tar -xf backup.tar -C backup-work
cd backup-work

jq -r '["\(.files_sha256)  files.tar.bz2.enc", "\(.hermes_sha256)  hermes.tar.bz2.enc"]
  + (if .users_sha256 then ["\(.users_sha256)  users.tar.bz2.enc"] else [] end)
  | .[]' manifest.json > SHA256SUMS
sha256sum -c SHA256SUMS

On macOS, replace the last command with shasum -a 256 -c SHA256SUMS. All listed payloads must report OK. Then enter the key without echoing it or placing it directly in shell history, decrypt, and extract:

read -rsp 'Backup decryption key: ' BACKUP_KEY; printf '\n'
printf '%s' "$BACKUP_KEY" > backup.key
unset BACKUP_KEY

openssl enc -d -aes-256-cbc -pbkdf2 -iter 200000 \
  -pass file:backup.key -in files.tar.bz2.enc | \
  tar -xjf - -C ../restored

openssl enc -d -aes-256-cbc -pbkdf2 -iter 200000 \
  -pass file:backup.key -in hermes.tar.bz2.enc | \
  tar -xjf - -C ../restored

if [ -f users.tar.bz2.enc ]; then
  mkdir -p ../restored/users
  openssl enc -d -aes-256-cbc -pbkdf2 -iter 200000 \
    -pass file:backup.key -in users.tar.bz2.enc | \
    tar -xjf - -C ../restored/users
fi

rm -f backup.key

Recovered customer files are under restored/webdav. Hermes configuration, conversations, skills, and persistent runtime state are under restored/home. Format 2 backups also place Hermes and administrator user-home state under restored/users. Format 1 backups created before user-home coverage remain restorable and contain the first two payloads only.

Why the download ends in .tar

The outer .tar is only a container for manifest.json and encrypted payloads. Each payload is compressed with bzip2 before AES-256 encryption (files.tar.bz2.enc, hermes.tar.bz2.enc, and, for format 2, users.tar.bz2.enc). Encrypted bytes are effectively incompressible, so compressing the outer archive would add work with almost no size reduction. The current format already gets the useful compression while keeping the manifest and payload boundaries easy to verify.

Overview

When a supported Hermes release is ready, you can start its upgrade from the Parlant Agents dashboard. Parlant does not replace the server until you choose to upgrade.

An upgrade creates and verifies a fresh encrypted backup, removes the old server, creates a replacement from the latest managed image, and restores the agent state. The hostname, secure tunnel, service credentials, storage, and subscription remain attached.

Before upgrading

• Wait for active agent work to finish.
• Review the release and any configuration changes you made.
• If the dashboard marks the VPS as customized, review its warning and choose whether registered scripts in $HERMES_HOME/vps-customizations/ should run again.
• Remember that one-off administrator changes are not replayed automatically.

Hermes is unavailable while the server is replaced. Follow the live upgrade steps and wait until the dashboard reports that the agent is ready before resuming work.

Upgrade, recreate, or restart?

Use Upgrade to move to the latest supported managed image. Use Recreate to replace the server with its current managed image version, still protected by a fresh backup. Use Restart for a simple power cycle when the server is unhealthy and replacement is unnecessary.

For backup coverage and recovery controls, see Backup and restore.

Introduction

Hermes is an autonomous agent from Nous Research. It can select tools, work across several steps, remember useful context, and create reusable skills from experience. Your managed instance stays online independently of your laptop.

Examples

Hermes is well suited to tasks that combine reasoning with tool use:

• research a subject, compare sources, and write a cited brief;
• inspect files, transform data, and produce a repeatable report;
• monitor a recurring question and deliver a scheduled summary;
• turn a successful multi-step process into a reusable skill.

A task is a poor fit when it requires an irreversible action that you cannot review, access to secrets the task does not need, or a result that cannot be independently checked.

In-depth guide

Start with the environment

Open Hermes from your Parlant Agents dashboard. Before assigning work, ask it to describe the active model, available tools, working directory, and any approval rules. This confirms what the current session can actually do.

Give complete requests

A useful request names the goal, available inputs, constraints, desired output, and stopping point. For example:

Compare the three documents in the project folder. Produce a Markdown table of conflicting claims, cite each file and page, and list unanswered questions. Do not edit the source files or contact anyone.

This is safer and easier to verify than “review these files.”

Review tool use

Hermes can use terminal, browser, search, file, and other configured tools. Keep approval controls enabled for consequential actions. Ask for a plan before a long task, and tell Hermes which actions require confirmation.

Build memory deliberately

Tell Hermes which stable preferences should persist, such as report format or preferred sources. Do not ask it to memorize temporary secrets. Start a fresh conversation when the subject changes substantially, and summarize any context that the new session still needs.

Reuse successful work

After a process works reliably, ask Hermes to describe the steps and edge cases before turning it into a skill. Test that skill on a small example. Reusable skills are most valuable when the inputs, output, and success criteria are clear.

For runtime-specific commands and integrations, consult the official Hermes documentation.

Introduction

Research is a strong first Hermes use case because the work can be divided into visible stages: define the question, find sources, extract evidence, compare claims, and write a deliverable. The final result remains reviewable by a person.

Examples

You might ask Hermes to compare software vendors, summarize recent regulatory changes, prepare background for an interview, or build a reading list for a technical decision.

A well-bounded prompt looks like this:

Research how three European hosting providers handle encrypted backups. Use primary documentation published or updated within the last 18 months. Produce a table covering retention, encryption, restore process, and price. Link every factual claim. Put uncertain or missing information in a separate section. Stop after writing hosting-comparison.md; do not contact vendors.

In-depth guide

1. Define acceptance criteria

Specify the geography, date range, source quality, fields to compare, output format, and prohibited actions. If freshness matters, require publication or update dates. If primary sources matter, say whether third-party summaries may be used only for discovery.

2. Ask for a research plan

Before tool use begins, ask Hermes to list likely source types and the evidence needed for each conclusion. Review the plan and remove irrelevant branches. This reduces wasted model credit and makes the final result easier to audit.

3. Separate evidence from interpretation

Ask Hermes to keep a source ledger with page title, URL, access date, quoted evidence, and the claim it supports. Require explicit labels for inference and uncertainty. A polished sentence without a supporting source should not be treated as established fact.

4. Produce a reviewable artifact

Markdown works well because it supports headings, tables, links, and version control. Ask for a short executive summary followed by detailed evidence. Keep raw notes in a separate file if you may need to revisit them.

5. Verify a sample

Open several links and compare the cited text with the brief. Check the claims that would most affect a decision, not only the easiest facts. If one citation is weak, ask Hermes to repair that claim and then check similar claims.

6. Improve the workflow

Once the output is consistently useful, save the prompt as a template or ask Hermes to create a skill. Keep variables such as subject, date range, comparison fields, and destination file explicit so that future runs remain predictable.

Introduction

Hermes includes scheduled automation and messaging integrations. Together, they can turn a tested one-off report into a daily or weekly briefing that runs while your laptop is offline.

Examples

Useful briefings include a weekday project-status digest, a weekly competitor update, a morning list of relevant support issues, or a recurring audit of public documentation links.

Start with a manual request:

Review the project notes changed during the last seven days. Summarize completed work, blocked work, and decisions needed next week. Link each item to its source file. Write the report only; do not send it yet.

Schedule it only after the manual result is accurate.

In-depth guide

1. Make one run reliable

Define the exact sources, time window, output format, and maximum length. Run the request manually several times with different inputs. Resolve missing-data and empty-result behavior before adding a schedule.

2. Choose a delivery channel

Hermes supports a messaging gateway for platforms including Telegram, Discord, Slack, WhatsApp, and Signal. Configure only the channel and recipients that need the report. Follow the official Hermes messaging guide for platform credentials and access controls.

3. Create the schedule

Tell Hermes the timezone, recurrence, request, and delivery destination. Ask it to repeat the interpreted schedule before saving it. Prefer an explicit timezone and time, such as “09:00 Europe/Berlin every weekday,” over “each morning.”

4. Define failure behavior

A scheduled task should report when a source is unavailable rather than silently inventing a complete result. Set a useful empty-state message, limit retries, and avoid sending partial reports as if they were final.

5. Observe the first runs

Check the first delivery time, source window, links, and recipient. Compare at least one scheduled result with a manual run. Review model-credit consumption in the Parlant Agents dashboard and adjust frequency or model choice if needed.

6. Pause before changing

Pause a schedule before changing its sources, credentials, or destination. Test the revised prompt manually, update the scheduled task, and confirm its interpreted recurrence again. Remove obsolete schedules so that they do not continue consuming model credit.

See the official Hermes cron guide and messaging guide for current runtime commands.

Introduction

You do not have to run Hermes on your phone to use it there. The simplest setup keeps Hermes on its always-on server and uses your phone as the remote control.

For most people, Telegram is the quickest and most reliable mobile option. It works on iPhone and Android, needs no open inbound port, supports files and voice messages, and can notify you when scheduled or background work finishes. A managed Parlant Agents dashboard can now create and connect a Telegram bot with a short QR flow.

Use a native app only when you need richer session, model, task, skill, memory, or file controls. Native apps and on-phone installs have different backends, so they are not interchangeable.

Examples

If you only want to send prompts and receive results, start with Telegram. It avoids almost all mobile networking and compatibility work.

In-depth guide

Fastest route for a managed agent: use the browser

1. Open your Parlant Agents account on the phone.
2. Select your agent and tap Hermes WebUI.
3. Sign in with the Hermex password shown under Credentials → Hermes WebUI and Hermex.
4. Open Chat and send a short test message.
5. Optional: in Safari use Share → Add to Home Screen; in Chrome use Menu → Add to Home screen.

This reaches the same agent, files, memory, and model configuration you use on a computer. Nothing runs on the phone beyond the browser.

Recommended on iPhone and Android: Telegram

1. Create and connect the bot with QR setup

1. Install Telegram and sign in.
2. Open your Hermes dashboard, then open Channels.
3. Find Telegram and select Set up with QR.
4. Scan the QR code or select Open Telegram. Telegram opens @HermesSetupBot.
5. Tap Create Bot. You may edit its display name before confirming.
6. Return to the Hermes dashboard. It should show the new bot username and detect your numeric Telegram user ID.
7. Check that your ID is in Allowed users, then select Save and restart.
8. Wait until Telegram shows Connected, open the new bot, and send Hello.

The setup service passes the new bot token back to your Hermes instance once. Hermes then stores and uses it from your own server.

2. Learn five useful commands

• /help — show available commands;
• /new — start a clean conversation;
• /stop — stop current work;
• /status — show session status;
• /sethome — send future scheduled results to this chat.

Telegram supports text, photos, documents, voice notes, streaming replies, and separate conversation topics. Start with a direct message; add groups or topics later.

3. Keep access private

Leave the allowlist enabled. Do not enable “allow all users” for an agent with terminal and file access. Treat the bot token as a password; if it leaks, revoke it through BotFather and reconnect the bot.

If QR setup is unavailable, use the manual path: create a bot with @BotFather, find your numeric ID with @userinfobot, then enter both under Channels → Telegram → Configure. The full fallback is in the official Telegram guide.

Native iOS route: Hermex

Hermex is a free, independent iOS client. The agent still runs on a computer or server. Hermex talks to the open-source hermes-webui API over HTTP or HTTPS and adds native sessions, streaming, attachments, model controls, tasks, skills, memory, and workspace browsing.

Important compatibility check: Hermex does not connect to the standard Hermes dashboard. Current managed Parlant Agents images run hermes-webui separately: https://<your-slug>.parlant.xyz is the WebUI and Hermex address, while https://<your-slug>-admin.parlant.xyz is the standard administration dashboard. Older managed instances show the legacy layout until upgraded.

1. Connect a managed Parlant Agents instance

1. Open your Parlant Agents account and select Credentials.
2. Copy Hermes WebUI URL and Hermex password from the Hermes WebUI and Hermex card.
3. Install Hermex from the App Store.
4. Paste the URL, tap Test Connection, enter the password, and sign in.

No server install, port forwarding, Tailscale, or extra Cloudflare account is needed. The service stays bound to loopback on your server and the existing outbound-only Cloudflare Tunnel supplies HTTPS. Parlant Agents pins and compatibility-tests hermes-webui with the Hermes version in each image and applies its managed theme automatically.

2. Install the required server on a DIY host

On a macOS, Linux, or Windows/WSL2 machine that already runs Hermes:

git clone https://github.com/nesquena/hermes-webui.git
cd hermes-webui
python3 bootstrap.py

The bootstrap checks for Hermes, creates its environment, and starts the web UI on 127.0.0.1:8787. Before making it reachable from another device, enable authentication in its Settings panel and set a strong, unique password. Confirm it works locally:

curl http://127.0.0.1:8787/health

3. Make a DIY installation reachable safely

Tailscale is the safer default. Install Tailscale on the server and iPhone, sign both into the same tailnet, then start hermes-webui with a password and a network-visible bind:

HERMES_WEBUI_HOST=0.0.0.0 HERMES_WEBUI_PASSWORD='use-a-long-unique-password' ./start.sh
tailscale ip -4

Use http://100.x.y.z:8787 in Hermex. Tailscale encrypts the connection. Binding to 0.0.0.0 also exposes the service to the local network, so keep the password enabled.

A stable HTTPS reverse proxy or named Cloudflare Tunnel also works. A quick tunnel is useful for testing:

cloudflared tunnel --url http://localhost:8787

Quick-tunnel URLs change after a restart. Any public hostname must use a strong server password. Hermex does not currently handle a separate Cloudflare Access login layer.

4. Connect the app

1. Follow the Hermex setup guide and install Hermex from the App Store.
2. Enter only the server origin, such as https://hermes.example.com or http://100.x.y.z:8787.
3. Tap Test Connection.
4. Enter the hermes-webui password and sign in.
5. Open an existing session or create a small test conversation.

If connection testing fails, check the server process, /health, Tailscale or tunnel status, then the URL and password—in that order. Hermex requires iOS 18 or later. Because its server API is still evolving, update Hermex and hermes-webui together when features stop working.

Android option 1: Play Store app for a separate on-phone Hermes

Hermes Agent - Android by Hen Works packages an agent, terminal, Python, git, sessions, memory, and provider setup into one Android app. Choose this when you want a new Hermes that runs on the phone itself.

1. Install the app from Google Play.
2. Create a separate provider API key with a low spend limit.
3. Follow the in-app setup, select a provider and model, then run its self-check.
4. Start with a harmless prompt before granting broader file or terminal access.

This app does not attach to your existing managed Hermes. It creates another installation with separate sessions, memory, and files. It is a third-party app with ads and an optional purchase, not an official Nous Research release. Review its Play Store privacy information and avoid reusing an unrestricted production API key.

Android option 2: community client for an existing DIY server

The Hermes Android community client can browse sessions and stream chat from an existing Hermes API server. Optional dashboard access adds memory, cron, skills, and settings. This is useful, but more work than Telegram.

1. On the Hermes host, enable the authenticated API server with a strong API_SERVER_KEY. Keep its default port 8642 private.
2. Put the server and Android phone on the same Tailscale network. Do not directly port-forward ports 8642 or 9119 to the public internet.
3. Download the current app-arm64-v8a-release.apk from GitHub Releases. Android 8 or later is required.
4. Allow APK installation for the browser or file manager only long enough to install it, then turn that permission off again.
5. In the app, add the server's Tailscale IP or MagicDNS name, port 8642, and the API key.
6. If you need dashboard panels, add the separately protected dashboard address and credentials.

This client is independent and sideloaded. Check the release page, repository activity, and requested permissions before every update. Managed Parlant Agents instances do not expose the API port by default, so this path needs server administration and is not the quick managed-service route.

Android option 3: official Hermes in Termux

Termux runs the real Hermes CLI inside Android. Use this when open-source code and local terminal control matter more than polish.

1. Install a current Termux build from termux.dev rather than an old, unsupported store build.
2. In Termux, run the official installer:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

3. Verify and configure it:

hermes version
hermes doctor
hermes model
hermes

The installer selects Android-compatible dependency groups automatically. The tested path includes the CLI, cron, background terminals, Telegram gateway, MCP, memory, and ACP. Docker isolation, local Whisper voice transcription, and automatic browser bootstrap are not part of the tested path. Android may suspend Termux in the background, so do not rely on it as an always-on gateway.

Termux is an official Tier 2 platform: real users run it, but releases can break Android-specific dependencies. Follow the complete Termux guide when installation or updates fail.

Which route should you choose?

• Managed Parlant Agents customer: browser first, Telegram next.
• Any user who wants reliable mobile chat and notifications: Telegram.
• iPhone user who wants a full native control surface and can operate an extra server: Hermex.
• Android user who wants a separate local agent with easy installation: Play Store app, with a limited provider key.
• Android user who wants a native client for an existing DIY server: community client over Tailscale.
• Android power user who wants official Hermes running on-device: Termux.

How these options were verified

Research checked current primary documentation, release activity, and real-user reports on 15 July 2026. Hermex had 42 US App Store ratings and reviews confirming daily use and straightforward setup; reviews also reported slow long conversations and confusion about the required hermes-webui sidecar. The Android Play Store app showed 10,000+ installs, a 4.4 rating from 2,016 ratings, and reviews confirming successful setup. The community Android client has active releases and a user report of a working reverse-proxy setup. Termux is documented and tested by the Hermes project, with active Android user reports that also confirm its Tier 2 limitations.

Introduction

A useful personal AI assistant does not begin with dozens of integrations or a perfect master prompt. It begins with one recurring job, a few clear boundaries, and an easy way to talk to it.

Hermes can search the web, work with files, remember selected preferences, and run multi-step tasks. Parlant Agents handles the server, secure web address, backups, and updates, so you can start without a terminal or code. You still choose what information to share and review every important result.

This guide takes you from a blank agent to a practical assistant brief and three safe first tasks.

Examples

Choose one job that already takes time each week:

Composite customer journey: from blank agent to useful routine

This is an illustrative journey built from common setup needs, not a testimonial from a named customer.

Mara runs a small design business. Her notes live across paper, messages, and half-finished lists, but she does not want another complex productivity system. She gives Hermes one job: turn Sunday’s rough notes into a plan she can approve.

On her first run, Hermes sorts the notes and spots that a client deadline conflicts with an appointment. Mara corrects one date and drops a low-value errand. She does not connect her inbox or calendar. After three useful manual runs, she asks Hermes to remember only two stable preferences—short checklists and one unscheduled evening—then connects her private Telegram bot for voice-note capture.

The success is not “full autonomy.” Mara gets a consistent starting point while every commitment, message, and account change stays under her control.

Good first work is easy to inspect and easy to undo. Avoid starting with “manage my whole life” or permission to send, buy, publish, delete, or book things by itself.

In-depth guide

1. Pick one useful job

Write one sentence before setup:

I want my assistant to turn my rough weekly notes into a short, realistic plan that I approve.

This gives you a visible success test. If the result does not save time or reduce confusion after a few attempts, improve the input or choose a better job instead of adding more tools.

2. Use the managed setup

Follow Getting started with Parlant Agents to create the instance and choose model access. Managed model credit is the shortest route when you do not already have a provider account. The monthly infrastructure subscription and model usage are separate costs; this billing guide explains both.

When the agent is ready, open Hermes from your account dashboard. No server command, DNS change, or software install is required.

3. Run a harmless capability check

Your tools depend on the current Hermes configuration. Ask what is actually available before designing a workflow:

Describe the tools available in this session in plain language. Do not use any tool or change anything. Tell me which three tools would be most useful for personal planning and research, and what each one cannot do.

This prevents an important mistake: assuming the assistant can see a calendar, inbox, or account that you never connected. Hermes only knows what you tell it, what exists in its permitted workspace, and what its configured tools can reach.

4. Your copy-and-paste assistant brief

Paste this into a new conversation and adjust the bracketed text:

Act as my personal planning and research assistant. Help me turn rough notes into clear options, next actions, and questions. Keep answers [short and practical]. When using the web, link factual claims and label uncertainty. Ask when a missing fact would change the answer; otherwise make a clearly labeled assumption. Never send a message, make a purchase or booking, publish, delete, share a file, or change an account without my explicit confirmation in the current conversation. Do not save sensitive personal details to memory. For now, only restate these rules as a five-item checklist and ask which task we should test first.

Correct the checklist until it matches what you meant. Then say:

Show me the short, stable preferences from that checklist that would be useful in future sessions. Do not save them yet.

Review the proposed text. Remove temporary details, private identifiers, and anything you do not want loaded into future conversations. Only then ask Hermes to save the approved preferences. Hermes memory is intentionally small and curated; use it for durable facts such as answer length or preferred plan format, not passwords, full schedules, or raw personal documents.

5. Test three low-risk tasks

Try one task from each group:

1. Organize: “Turn these rough notes into Do next, Waiting for, and Questions. Do not invent dates.”
2. Draft: “Draft a polite reply from these points. Do not send it. Mark any fact you had to assume.”
3. Compare: “Compare these three options against my criteria. Use current primary sources, link each material claim, and leave unknowns blank.”

A good result should be specific, reviewable, and within the requested boundary. If it misses, correct one thing at a time: input, output format, source quality, or stopping point. Do not solve a vague prompt by granting broader access.

6. Put it where you will use it

The browser is enough to begin. For quick capture on a phone, connect the Telegram bot after the browser workflow works. You can send text, files, photos, and voice notes from iPhone or Android. Hermes on iPhone and Android walks through the managed QR setup and access allowlist.

Keep the bot private. Only your user ID should be allowed unless you deliberately configure another person. Treat bot tokens and provider keys like passwords.

7. Know the privacy boundary

A dedicated managed instance isolates your Hermes runtime and files from other customers, but model and tool providers receive the content needed to answer a request. Share the minimum useful excerpt. Remove account numbers, identity documents, passwords, private keys, and unrelated personal details before pasting or uploading material.

Review high-impact advice independently. An AI-produced summary can miss a deadline or misread a clause; it is not a replacement for a doctor, lawyer, accountant, or other qualified professional.

8. Add automation last

Keep the workflow manual until it produces a useful result at least three times with different inputs. Then you can ask Hermes to turn the steps into a reusable skill or create a scheduled reminder. Keep purchase, booking, account, and communication actions behind explicit approval.

For feature details, see the official Hermes guides to best practices, persistent memory, and available capabilities.

Introduction

Weekly planning is a strong personal AI assistant use case: the input is messy, but the output is simple to inspect. You bring the fixed commitments, unfinished tasks, errands, and worries. Hermes sorts them, finds conflicts, and proposes a week you can change before anything happens.

This workflow does not require calendar automation. Paste only the details needed for the coming week, or send a voice note through Telegram. Hermes first extracts what you said, then builds the plan after you correct it. That separation catches mistakes before they become a schedule.

Examples

A useful weekly reset can turn this:

Dentist Tuesday 15:00. Finish the insurance form. Call Sam. Need groceries before guests arrive Friday. Presentation is important but I keep putting it off. Wednesday looks packed. I would like one evening with no plans.

into this:

• three outcomes that would make the week successful;
• a table of fixed commitments and deadlines;
• suggested focus blocks around those commitments;
• errands grouped into one batch;
• a short “waiting for” list;
• tasks to defer, delegate, or drop;
• conflicts and missing information for you to resolve.

Composite customer scenario: the overbooked freelancer

This scenario illustrates the workflow; it is not presented as a real customer testimonial.

Jonas, a freelance photographer, starts with two client deadlines, an insurance form, a family visit, and a promise to keep Thursday evening free. He sends a voice note instead of preparing a perfect task list. Hermes first extracts the commitments and asks how long Tuesday’s appointment lasts. Only after Jonas answers does it build the week.

The first draft shows that every task cannot fit without removing all buffer. Jonas defers a portfolio refresh, batches two errands after the appointment, and protects one morning for the harder client edit. When a delivery moves on Wednesday, Hermes changes only the affected day rather than rewriting the whole week.

The useful outcome is a visible trade-off: one task deliberately moves, instead of five tasks quietly slipping.

The same method works for a busy parent coordinating household tasks, a freelancer balancing clients and admin, a job seeker tracking applications, or anyone who feels their task list is larger than their week.

In-depth guide

1. Gather only this week’s inputs

Bring five small lists:

1. Fixed: appointments, events, and real deadlines.
2. Important: up to three outcomes that matter most.
3. Loose: errands, calls, chores, and ideas.
4. Waiting: people or information you cannot proceed without.
5. Capacity: unavailable times, travel, health constraints, and how much unscheduled space you want.

Hermes does not automatically know your calendar, inbox, or family commitments unless you explicitly provide or configure access. Paste a compact list instead of an entire inbox or calendar export. Redact addresses, booking references, account numbers, and details unrelated to planning.

If typing the list feels like work, send yourself a two-minute voice note through your private Telegram bot. Hermes can transcribe incoming voice messages. Read the transcript before using it as planning input, especially for names, dates, and times.

2. Sort before planning

Start a fresh conversation and paste or dictate the lists with this prompt:

This is raw input for my weekly reset. Do not build a schedule yet. Extract only what I explicitly said into a table with Item, Type, Date or deadline, Estimated effort, and Missing information. Use unknown instead of guessing. Then list possible conflicts and ask no more than five questions that would materially change the plan. Do not contact anyone or change any external system.

Correct transcription errors and assumptions. Add only missing information that affects the week. This step is deliberately boring: clean inputs make the next step useful.

3. Define a realistic week

Tell Hermes your normal limits in plain language. For example:

Plan at most two demanding tasks per day. Leave 30 minutes between appointments. Keep Wednesday evening and at least one other evening free. Treat estimates as ranges, not promises.

These are good candidates for memory if they stay true most weeks. Ask Hermes to show the exact memory entry first, then approve it. Keep temporary appointments and sensitive details out of persistent memory.

4. Build the plan

Once the extracted list is correct, use this prompt:

Build a realistic plan for [date] through [date]. Preserve every fixed commitment. Choose at most three weekly outcomes and explain why. Place demanding work in my available focus periods, batch compatible errands, and keep the requested free time. Do not fill every open hour. Put anything that does not fit under Defer, delegate, or drop. Finish with Monday’s first 15 minutes and a short checklist. This is a proposal only; do not change a calendar, send messages, make bookings, or purchase anything.

Ask for a table if you scan schedules visually, or a daily checklist if tables feel heavy. One format is enough; extra versions consume time and model credit without improving the decision.

5. Make Hermes challenge the plan

Before accepting it, ask:

Stress-test this plan. Find double-booking, hidden travel or preparation time, vague tasks, unrealistic effort, and days with no buffer. Name the two tasks most likely to slip. Suggest the smallest change that makes the week more robust.

Then make the final decisions yourself. A plan should show trade-offs instead of quietly squeezing everything in.

6. Save one working artifact

Ask Hermes to write the approved version to a clearly named file such as weekly-plan-2026-07-20.md. Keep these sections:

• Top three outcomes
• Fixed commitments
• Daily plan
• Waiting for
• Defer, delegate, or drop
• First action Monday

Do not use the weekly plan as a second password vault, medical record, or identity folder. Keep only what helps execute the week.

7. Use a two-minute daily check-in

Each morning, reply in the same conversation or open the plan and say:

Today changed: [new fact]. Show only what this affects. Keep fixed commitments and free-time boundaries. Give me the next physical action, not a new full-week plan.

Each evening, mark completed, moved, and blocked items. Avoid rebuilding the entire schedule for every interruption.

8. Close the loop

At week’s end, ask three questions:

1. What was completed?
2. What repeatedly slipped, and why?
3. Which one stable preference or workflow lesson is worth remembering?

Approve memory updates individually. A useful lesson is “morning focus blocks work better than evening blocks”; a poor memory entry is a complete diary of the week.

After several successful manual resets, schedule a plain reminder such as “Send me my weekly-reset checklist every Sunday at 18:00 Europe/Berlin.” Automate source gathering only when each source is reliable and the output remains reviewable. Follow the scheduled briefings guide before turning the planning prompt into a recurring task.

For current platform behavior, see the official Hermes documentation for memory, scheduled tasks, and Telegram voice and file messages.

Introduction

Parlant Agents runs Hermes for you on dedicated infrastructure. You do not need to configure a virtual server, DNS, TLS certificates, or backup scripts. This guide takes you from account creation to your first conversation.

Examples

A new customer can use the service in either of these ways:

• Managed model credit: buy a fixed amount of OpenRouter credit during setup. Parlant Agents installs a restricted key and shows the remaining balance in your account dashboard.
• Your own provider key: provision the agent first, then add a Hermes-supported provider key in the Hermes dashboard. Your provider bills you directly.

In both cases, your monthly Parlant Agents subscription pays for the managed server, private storage, hostname, TLS, and backups. Model usage is separate.

In-depth guide

1. Create your agent

Select Create agent, choose Hermes, and enter your email address. Parlant Agents sends a secure magic link to verify that address. Open the link in the same browser to continue.

2. Review the subscription

The review page shows the current monthly infrastructure price and your future agent address. Stripe handles payment securely. The subscription renews monthly until you cancel it through the billing portal.

3. Select model access

Choose managed OpenRouter credit for the quickest setup, or choose Use your own key if you already have a supported provider account. Managed credit packs do not renew automatically. You can add credit later from the Parlant Agents dashboard.

4. Follow provisioning

The progress page reports real infrastructure work: credentials, server and storage, network, DNS, TLS, model access, and the final health check. You can safely leave the page and return to your account dashboard.

5. Open Hermes

When provisioning finishes, open your Parlant Agents dashboard and select the agent. Copy the Hermes dashboard credentials or use the Open agent link. Sign in to Hermes, confirm the selected model, and begin with a bounded request such as:

Summarize what you can do in this environment. Do not make any changes. Suggest three useful first tasks and explain which tools each task would use.

Keep your first tasks narrow. Review tool requests and results before giving Hermes broader or recurring work.

Introduction

A Parlant Agents subscription and language-model usage are two different costs. The subscription keeps your dedicated agent infrastructure online. Model access pays the provider that answers prompts and decides which tools Hermes should use.

Examples

Managed credit fits someone who wants to start without opening a separate model-provider account. A $10 credit pack adds $10 to a restricted OpenRouter key. Stripe displays the separate Parlant Agents service fee before payment.

Bring your own key fits a team that already has an OpenRouter, Anthropic, OpenAI, or other Hermes-supported provider account. The provider charges that account directly, and Parlant Agents does not add a model-credit service fee.

In-depth guide

Monthly subscription

The live infrastructure price appears before checkout. It covers dedicated compute and storage, the managed hostname and TLS connection, provisioning, and encrypted backups. It renews monthly until canceled.

Managed OpenRouter credit

During setup, select one of the available fixed credit packs. Credit never renews automatically. Parlant Agents creates a restricted child key, applies exactly the purchased credit, and reports usage and the remaining balance in your dashboard.

OpenRouter reports costs in US dollars. Token quantity varies by model and by the mix of input and output tokens. A stronger model can consume the same credit faster than a smaller model.

To add credit later:

1. Open Dashboard and select your agent.
2. Open the Credits panel.
3. Select a credit pack.
4. Review both the credit amount and service fee in Stripe before paying.
5. Return to the dashboard and wait for the balance to update.

Your own provider key

Select Use your own key during setup. Once Hermes is ready, open its dashboard, add the provider secret under Keys, and select an available model under Models. Provider capabilities and prices come from that provider.

Treat API keys like passwords. Give a key only the permissions and spending limit the agent needs. Revoke and replace it in the provider account if it is exposed.

Choosing between them

Start with managed credit when convenience and a fixed initial budget matter most. Bring your own key when you need direct provider billing, organization-level controls, or a provider that is not available through managed OpenRouter access. You can later switch an existing instance from your own key to managed credit in the Parlant Agents dashboard.