HostedGPT is a free, open-source alternative to ChatGPT. It's a Ruby on Rails app so you can run it on any server or even your own computer. Just bring your own OpenAI API key.
This app is designed to be incredibly easy for ChatGPT users to switch. All the features you expect are here plus it supports Claude 3 and GPT-4 in a single app. You can also switch assistants in the middle of a conversation!
This project is led by an experienced rails developer, but I'm actively looking for contributors to help!
For the easiest way to get started, deploy a full version of HostedGPT to the hosting service, Render, for free. This free app works for 90 days and then the database will stop working. You will need to upgrade to a paid version of the database which is $7 / month. Alternatively, you can also run it off your local computer. Jump down to the Developer Instructions if you want to run it locally.
Click Fork > Create New Fork at the top of this repository
Create an account on Render.com and login. If you are new to Render, you may be prompted to add a credit card to your account. However, you will be on their free plan by default unless you choose to upgrade.
View your newly created fork within github.com and click the button below (be sure you're viewing your fork of this repo before clicking):
In Blueprint Name, type anything such as "hostedgpt-
Click Apply
Wait for the hostedgpt database and web service to be deployed. The first deploy takes a good 5 minutes so be patient. After they are all deployed, click Dashboard at the top of the Render screen. Or if an error occurs, skip to Troubleshooting Render below.
You should see two "Service Names" called "hostedgpt-..." (the name you picked), click the one that is of type Web Service
On the details screen, click the URL that looks something like hostedgpt-XXX.onrender.com
NOTE: After 15 minutes of not using the app your Render server will pause. Next time you visit the first request will auto-resume the server, but this resume is slow. If this annoys you, upgrade Render for $7 per month:
You may want to read about configuring optional features.
If you encountered an error while waiting for the services to be deployed on Render:
Deploying to Fly.io is another great option. It's not quite one-click like Render and it's not 100% free. But we've made the configuration really easy for you and the cost should be about $2 per month, and Render costs $7 per month after 90 days of free service so Fly is actually less expensive over the long term.
brew install flyctl
otherwise curl -L https://fly.io/install.sh | sh
(view instructions)fly launch --build-only --copy-config --name=APP_NAME_YOU_CHOSE
Fly Postgres
with a unique name such as [APP_NAME]-db
and you can set the configuration to Development
.Confirm Settings
at the bottom of the page and close the browser.bin/rails db:setup_encryption[true]
. This will initialize some private keys for your app and send them to Fly.fly deploy --ha=false
Development
as the DB size in the step above, now you should run bin/rails db:fly[APP_NAME_FROM_EARLIER,swap,512]
This will increase the swap on your database machine so that it doesn't crash.You may want to read about configuring optional features.
Heroku is a one-click option that will cost $10/month for the compute (dyno) and database. By default, apps use Eco dynos ($5) if you are subscribed to Eco. Otherwise, it defaults to Basic dynos ($7). The Eco dynos plan is shared across all Eco dynos in your account and is recommended if you plan on deploying many small apps to Heroku. Eco dynos "sleep" after 30 minutes of inactivity and take a few seconds to wake up. Basic dynos do not sleep.
Eligible students can apply for Heroku platform credits through Heroku for GitHub Students program.
View your newly created fork within github.com and click the button below (be sure you're viewing your fork of this repo before clicking):
You may want to read about configuring optional features.
The file options.yml
contains a number of Features and Settings you can configure. Simple flags are:
REGISTRATON_FEATURE
is true
by default but you can set to false
to prevent any new people from creating an account.VOICE_FEATURE
- This is an experimental feature to have spoken conversation with your assistant. It's still a bit buggy but it's coming along.GOOGLE_TOOLS_FEATURE
— This is an experimental feature that enables your assistant to access your Gmail (and soon Google Tasks and Calendar).You first need to follow all the steps in the Google OAuth instructions. The only step that is optional is that you can leave GOOGLE_AUTHENTICATION_FEATURE
set to false, which means you don't have to enable new users to register with Google. However, following all the steps will also set up Google Auth so you can connect Google Tools to your assistants. After, you complete those steps, here is the additional configuration you need to do in order to enable the Google tools:
Go back to the OAuth Consent Screen:
Finally, set GOOGLE_TOOLS_FEATURE
to true
HostedGPT supports multiple authentication methods:
Password authentication is enabled by default. You can disable it by setting PASSWORD_AUTHENTICATION_FEATURE
to false
.
Google OAuth authentication is disabled by default. You can enable it by setting GOOGLE_AUTHENTICATION_FEATURE
to true
.
To enable Google OAuth authentication, you need to set up Google OAuth in the Google Cloud Console. It's a bit involved but we've outlined the steps below. After you follow these steps you will set the following environment variables:
GOOGLE_AUTH_CLIENT_ID
- Google OAuth client ID (alternatively, you can add google_auth_client_id
to your Rails credentials file)GOOGLE_AUTH_CLIENT_SECRET
- Google OAuth client secret (alternatively, you can add google_auth_client_secret
to your Rails credentials file)Steps to set up:
Go to the Google Cloud Console and Create a New Project:
Create OAuth Consent Screen:
Create OAuth Credentials:
https://example.com
https://example.com/auth/google/callback
https://example.com/auth/gmail/callback
Set Environment Variables:
GOOGLE_AUTH_CLIENT_ID
: Your Client ID ENV or google_auth_client_id
in your Rails credentials fileGOOGLE_AUTH_CLIENT_SECRET
: Your Client Secret ENV or google_auth_client_secret
in your Rails credentials fileHTTP header authentication is an alternative method to authenticate users based on custom HTTP request headers. This method is useful when you have an existing authentication system, and you want to direct users to HostedGPT and have them skip all authentication steps. They'll be taken right into the app and a HostedGPT user account will be created on the fly. This works by having your existing system set custom headers for authenticated users. This may be a Reverse Proxy (e.g., Traefik or Caddy) or a Zero Trust Network (e.g., Tailscale).
Steps to set up:
Enable the feature:
HTTP_HEADER_AUTHENTICATION_FEATURE
to true
. This will automatically disable password and Google OAuth authentication methods.Configure the request headers:
HTTP_HEADER_AUTH_EMAIL
- Set this environment variable to the name of the HTTP request header which will contain the user's email address. This defaults to a check for a request header of X-WEBAUTH-EMAIL
to find the user's email.HTTP_HEADER_AUTH_NAME
- Set this environment variable to the name of the HTTP request header which will contain the user's full name (first and last). This defaults to a check for a request header of X-WEBAUTH-NAME
to find the user's full name.HTTP_HEADER_AUTH_UID
- Set this environment variable to the name of the HTTP request header which will contain the user's identifier (any unique alphanumeric string). This defaults to a check for a request header of X-WEBAUTH-USER
to find the user's unique ID.Test the connection:
We welcome contributors! After you get your development environment setup, review the list of Issues. We organize the issues into Milestones and are currently working on v0.7. View 0.7 Milestone. Look for any issues tagged with Good first issue and add a comment so we know you're working on it.
The easiest way to get up and running is to use the provided docker compose workflow. The only things you need installed on your computer are Docker and Git.
git clone [repository url]
cd
into your clonedocker compose up --build
to start the appdocker compose run base rails test
The app has comprehensive test coverage but note that system tests currently do not work in docker.docker compose run base rails console
docker compose run base psql
.editorconfig
file to help eliminate whitespace differences in pull requests. It's nice if you install an extension in your IDE to utilize this (e.g. VS Code has "EditorConfig for VS Code").Every time you pull new changes down, kill docker (if it's running) and re-run:
docker compose up --build
This will ensure your local app picks up changes to Gemfile, migrations, and docker config.
HostedGPT requires these services to be running:
brew install imagemagick
should work on Mac )cd
into your local repository clonerbenv install
to install the correct ruby version (it reads the .ruby-version in the repo)bin/dev
starts up all the services, installs gems, and inits database (don't run db:setup as it will not configure encryption properly)bin/rails test
and bin/rails test:system
to run the comprehensive tests.editorconfig
file to help eliminate whitespace differences in pull requests. It's nice if you install an extension in your IDE to utilize this (e.g. VS Code has "EditorConfig for VS Code").Every time you pull new changes down, kill bin/dev
and then re-run it. This will ensure your local app picks up changes to Gemfile and migrations.
If you're set up with Docker you run docker compose run base rails test
. Note that the system tests, which use a headless browser, are not able to run in Docker. They will be run automatically for you if you create a Pull Request against the project.
If you set up the app outside of Docker, then run the usual bin/rails test
and bin/rails test:system
.
The Dockerfile
is set up to support three distinct situations: development, deploying to Render, and deploying to Fly. Each of these are completely separate targets which don't share any steps, they are simply in the same Dockerfile.
The docker-compose.yml
is solely for development. It references the development
build target.
The render.yml
specifies details of the Render production environment. Note that Render does not support specifying a build target within this file, it simply defaults to the last target with the Dockerfile so the order of the sections within there matter.
The fly.toml
specifies details of the Fly production environment. It references the fly-production
build target. The Fly section of the Dockerfile was generated using the dockerfile-rails generator. This is Fly's recommendation and it produces a reasonable production-ready Dockerfile. Edits to this top section of the file have been kept very minimal, on purpose, because it's intended to be updated using the generator. When it was originally generated it saved all the configuration parameters into config/dockerfile.yml
. When you run bin/rails generate dockerfile
it will read all these configurations and attempt to re-generate the Dockerfile. You can try this, it will warn you that it's going to overwrite, and press d
to see the diff of what changes it will make. There should be no functional changes above the line #### END of FLY ####
. Imagine you wanted to use this generator to change the app to use MySQL (view all generator options). You could run bin/rails generate dockerfile --mysql
and it would update your Gemfile, automatically run bundle install to install any gem changes, and then it will attempt to update Dockerfile where you can again press d
. Inspect the diff of any changes above the line #### END of FLY ####
and manually apply those changes. Similarly, view the diff for dockerignore and docker-entrypoint, although none of those changes should be necessary. When you get to fly.toml
you will want to view that diff closely and manually apply those changes. At the end it will update config/dockerfile.yml to record the new configuration of the Dockerfile. In this way, you can continue to use the generator to keep the Dockerfile updated (as recommended by Fly) while not breaking the dev or Render setup.
(Notable features being developed for v0.7: Heroku deploy, voice support, skills for the AI, Gemini Pro, pin conversations)
v0.6 - Released on 4/26/2024
v0.5 - Released on 2/14/2024