chatGPT-discord-bot

Migration from the legacy bot

This configuration-breaking modernization uses package version 4.0.0. The existing v3.0.0 Git tag identifies the legacy code and is preserved; a branch push does not create a new release or move an existing tag.

Stop the old bot and preserve .env, custom prompts and any user-maintained files. The old implementation held a single shared history in memory; it had no safe per-user durable database to import. Do not assign that shared history or arbitrary “latest” native sessions to a user. The new database starts empty.

1. Create a Python 3.12–3.14 virtual environment and install the new requirements. Do not reuse the old dependency environment. Browser automation, cookie login, unofficial endpoint aggregators and their hidden fallbacks have been removed.
2. Copy config.example.toml to config.toml, and create .env using the example. Keep DISCORD_BOT_TOKEN. Convert API variables using the table below. Set the default model alias explicitly; the example uses a local server.
3. Configure allowed/admin user IDs and channels in TOML. Use a real local model ID or your official API’s current ID. Copy any customized legacy system_prompt.txt text into bot.system_prompt; it is now an instruction, not a paid startup conversation. The unused placeholder is no longer shipped. Preserve your own prompt files before upgrading; the new bot does not read them.
4. Remove outdated settings from the active .env. Preserve any archive privately. Startup rejects obsolete settings instead of silently changing backend or billing behavior. Consumer-site cookies and shared free credentials cannot be configured as integrations. Official native CLI account login is a separate explicit mode described below.
5. Start the bot once; it synchronizes the current slash commands. Test two users, two channels, private/public mode, a restart and a reset before wider use.

Previous configuration	New configuration
OPENAI_KEY	OPENAI_API_KEY or another variable referenced by api_key_env
CLAUDE_KEY	ANTHROPIC_API_KEY
GEMINI_KEY	GEMINI_API_KEY
GROK_KEY	XAI_API_KEY
DEFAULT_PROVIDER, DEFAULT_MODEL, MODEL	TOML backend/model aliases and bot.default_model
ADMIN_USER_IDS	bot.admin_user_ids = [123, 456]
MAX_CONVERSATION_LENGTH, CONVERSATION_TRIM_SIZE	max_turns, context_bytes, retention_days
REPLYING_ALL, REPLYING_ALL_DISCORD_CHANNEL_ID	bot.reply_channels plus admin /replyall enabled
DISCORD_CHANNEL_ID startup response	No automatic startup inference/message
LOGGING	Safe operational logging always enabled; no prompt logs
OPENAI_ENABLED, OPENAI_TEMPERTUARES	Explicit alias and supported parameters
BING_COOKIE, GOOGLE_PSID, G4F_*, free/g4f provider kind	Removed; choose an official API or administrator-hosted compatible server

Changes users should expect:

• Private and public histories are separate, and default slash replies are private. Provider/persona/reset actions affect only the user’s current context.
• Long responses arrive as text files rather than many public follow-up messages.
• Provider/persona changes preserve bot text history and rebuild native context.
• /draw and /video require a media alias; an unavailable media provider cannot trigger another paid service. Existing unsupported media code was removed.
• Official APIs with legitimate free tiers remain supported through API keys and the same explicit billing/account choices.

Schema version 1 is newly introduced. Existing unknown schemas are rejected before schema creation; do not overwrite a database to make startup succeed. Store data in a private directory owned by the bot, and mount it persistently in Docker. Use SQLite backup or stop the bot before copying SQLite/WAL files. Never run two processes against the same database. Rollback means stopping the new bot and using the old code/config in a separate environment; do not interpret a new database as legacy data.

Opt into native CLI account login

Existing API/local configurations and schema version 1 remain valid. To enable subscription/account mode, merge the needed blocks from config.account.example.toml and follow CLI provisioning and login. Restrict the entire bot to the account owner, rebuild pinned CLI images with the account helper, and configure the account proxy. Use a fresh private auth_profile metadata directory for each backend. Do not import an existing personal CLI credential cache or overwrite the database.

Remove api_key_env from account backends; they reject mixed authentication. Keep separately configured API/media keys where needed. Native browser/device login replaces token environment variables for account mode. Switching modes or re-logging in preserves bot history and reconstructs native context. /reset and retention keep account login; /cli_auth action:logout model:ALIAS or local cli-auth logout ALIAS signs out. Codex/Grok device login can now be initiated through the always-private /cli_auth command; Claude login remains in the local terminal. No additional OAuth token environment variable or database migration is needed. Shared subscription bots remain unsupported; CLI media is not exposed by this bot (see native capability boundaries).

Image search and xAI video aliases

No schema migration or credential replacement is needed. Merge the optional models.image_search and models.xai_video blocks from the main example, pointing them at your configured xAI API backend. Restart to register /image_search; use /video model:xai_video explicitly for xAI video. Existing OpenAI image and Veo aliases remain valid. This adds no automatic fallback or recurring paid tests.

This site is open source. Improve this page.