Besidka: open-source BYOK AI chat
Besidka is an open-source BYOK (bring your own key) AI chat app. Pay the AI provider only for what you use, see the price of every message, and use all the models in one chat. No subscription, no ads, no markup. Self-hostable.
Who it's for
Besidka suits people who use AI every day for search, translation, PDFs, images or code, do not want another subscription, and are fine creating an API key in a provider's developer console. It is probably not for you if you already pay for ChatGPT Plus or Pro, Claude Pro or Google AI Pro and rarely hit their limits, since those apps do more than a chat window, or if you would rather not deal with API keys, or with prepaying credit at a provider. For scale, the maintainer's own non-work use costs between $1 and $5 a month. That is one person's usage, not a promise.
Get started
- Sign up at besidka.com with email and password, Google or GitHub.
- Add an API key at besidka.com/profile/keys. The easiest start is a free Gemini key from Google AI Studio, which needs no card. On Google's free tier your prompts may be used to improve its products, so keep sensitive material out.
- Start a new chat at besidka.com/chats/new and pick a model.
New to API keys? Read the BYOK guide.
Project
Project board is available here.
Contributing
Issues are welcome. Code contributions are by invitation only; see
CONTRIBUTING.md.
Tech stack
Features
- PINK: Dark and light themes, both unapologetically pink
- BYOK (Bring Your Own API Key): Use your API keys from any supported LLM provider (Anthropic, Google, OpenAI, xAI, DeepSeek, Moonshot AI, Qwen), route through gateways (Vercel, Cloudflare, OpenRouter), and search with Brave or Exa. Only pay for what you use. No subscriptions, no markup
- Multiple AI Models: Switch between the latest models from all providers in one place
- Web Search: Ground AI answers with real-time web context; use native provider search, Brave, or Exa
- Reasoning Mode: Step-by-step thinking for complex questions
- AI Image Generation: Generate one image per request with supported OpenAI and Google models, then preview or download it from your private Generated by AI file library; generated images count toward your storage quota
- Deep Research: Dedicated research-agent models browse the web and write cited reports; runs asynchronously with a push notification when it's ready, billed per-task on your own API key
- File Attachments & File Manager: Send images and PDFs, reuse files across chats
- Chat Branching: Fork a conversation mid-thread to explore a different direction
- Shareable Chats: Share a chat via a public, unguessable link that anyone can view without an account and branch into their own chats; control link expiry, search-engine indexing, and what's visible per share
- Projects: Organize chats with shared instructions and memory
- Resumable Streams: Refresh the page and pick up right where you left off
- Chat History: Search, pin and rename past conversations
- Syntax Highlighting: Clean, readable code blocks out of the box
- Mermaid Diagrams: Flowcharts, sequence, class, state, ER, Gantt, pie and more render as themed SVG diagrams right in the chat, with a one-click toggle back to the source
- No Install Required: Works in any browser, sign in and go
- Better Auth: Email/password, Google or GitHub sign-in
OAuth troubleshooting
Social OAuth in in-app browsers
- Symptom: Google or GitHub social sign-in is unavailable, or Google shows
403: disallowed_useragent.
- Cause: The app is opened in an embedded/in-app browser
(for example Threads, Instagram, Facebook, TikTok), and OAuth providers
may block these user agents.
- Resolution: Open the same page in your system browser
(Chrome/Safari/Firefox) and try again with social sign-in.
- Config sanity check: Ensure your Google OAuth app has this redirect URI:
https://<your-domain>/api/auth/callback/google
Local installation
Pay your attention that the project is designed to run on Cloudflare Workers. It requires additional steps to run it via Cloudflare Workers preview or deploy to the production environment.
The steps below are for local development only for the quick start and check.
Prerequisites
Steps
Clone the repository.
git clone [email protected]:besidka/besidka.git
Go to the project directory. Install the dependencies.
cd besidka
pnpm install
Copy ENV related files.
cp .dev.vars.example .dev.vars
Generate environment types for the project.
pnpm run cf-typegen
Prepare drizzle migrations.
# Expected output is .drizzle/migrations/*.sql
pnpm run db:generate
Apply the migrations to the D1 database.
# Expected output is .wrangler/state/v3/d1/*.sqlite
pnpx wrangler d1 migrations apply DB
Prepare and apply the cookie-consent database migrations. This is a separate
D1 database (CONSENT_DB) used to store consent receipts. See
docs/cookie-consent.md for details.
# Expected output is .drizzle/migrations-consent/*.sql
pnpm run db:consents:generate
pnpm run db:consents:migrate
Start the development server.
pnpm run dev
or to start Cloudflare Workers preview server
pnpm run preview
- Open http://localhost:3000 in your browser.
- Sign up http://localhost:3000/signup. Please use the Email + Password flow because you don't have prepared API keys for Google and GitHub OAuth yet. In development mode you don't need to wait for email confirmation. You have to be automatically redirected to the home page as a customer already.
- Put your own API keys here: http://localhost:3000/profile/keys
- You are welcome to start a new chat: http://localhost:3000/chats/new
Content (Nuxt Content + Studio)
The landing page is powered by Nuxt Content v3 with a dedicated Cloudflare D1 database (CONTENT_DB) that is fully isolated from the main application database (DB). Nuxt Studio provides an optional in-app visual editor that commits changes directly to the GitHub repository.
Creating the D1 databases
Run these commands once after cloning. They print a database_id UUID — copy it into the matching "TO_BE_CREATED" placeholder in wrangler.jsonc.
# Preview environment (default env in wrangler.jsonc)
pnpm exec wrangler d1 create besidka-content
# Production environment
pnpm exec wrangler d1 create besidka-content --env production
Replace both "TO_BE_CREATED" values in wrangler.jsonc under the CONTENT_DB binding with the returned UUIDs.
Nuxt Content manages its own schema inside CONTENT_DB at runtime — no Drizzle migrations are needed for this database.
Studio OAuth setup
Studio is optional. The landing page renders and all content is editable via Git without it. Studio adds a /_studio route to the deployed site so editors can make changes through a browser UI.
- Go to GitHub → Settings → Developer settings → OAuth Apps → New OAuth App.
- Set Homepage URL to
https://besidka.com (the canonical apex host — a Cloudflare redirect sends www.besidka.com to besidka.com).
- Set Authorization callback URL to
https://besidka.com/_studio (production) or http://localhost:3000/_studio (development).
- Copy the Client ID and generate a Client Secret.
- Add them to
.dev.vars for local development:
STUDIO_GITHUB_CLIENT_ID=<your_client_id>
STUDIO_GITHUB_CLIENT_SECRET=<your_client_secret>
- Add them as Wrangler secrets for the deployed Worker:
pnpm exec wrangler secret put STUDIO_GITHUB_CLIENT_ID
pnpm exec wrangler secret put STUDIO_GITHUB_CLIENT_SECRET
# For production:
pnpm exec wrangler secret put STUDIO_GITHUB_CLIENT_ID --env production
pnpm exec wrangler secret put STUDIO_GITHUB_CLIENT_SECRET --env production
Landing page without Studio
The landing page works without Studio configured. Content lives in content/landing/ as Markdown files and is committed to the repository like any other source file. Studio is purely an authoring convenience — it does not affect how the landing page is built or served.
Security
Snyk code checking repository
- Snyk is a tool for finding and fixing vulnerabilities in your code.
# Total issues: 0
snyk code test
Preview
