Add initial llm-bridge project

providers/chatgpt.toml

@@ -0,0 +1,86 @@
+# ChatGPT provider configuration, driven entirely by this file.
+# Selectors change when the web UI is updated: edit them here (reloaded live),
+# never in the Rust source. Validate with: llm-gateway selftest chatgpt
+
+[provider]
+name = "chatgpt"
+display_name = "OpenAI ChatGPT"
+web_url = "https://chatgpt.com/"
+new_conversation_url = "https://chatgpt.com/"
+# Captures the web conversation id from the current URL (one capture group).
+conversation_url_pattern = "^https://chatgpt\\.com/c/([0-9a-zA-Z-]{8,})"
+default_models = ["gpt-4o", "gpt-4o-mini", "gpt-4.1"]
+enabled = true
+
+[selectors]
+input_field = "div#prompt-textarea, div.ProseMirror[contenteditable='true']"
+# The id and the data-testid are language independent; the aria-label fallbacks
+# only match an English UI (see providers/selectors.md, "Localised user
+# interfaces").
+send_button = "button[data-testid='send-button'], button#composer-submit-button, button[aria-label='Send prompt'], button[aria-label='Envoyer la requête']"
+response_container = "div[data-message-author-role='assistant']"
+streaming_indicator = "button[data-testid='stop-button']"
+file_input = "input[type='file']"
+new_chat_button = "a[data-testid='create-new-chat-button']"
+# Clicked once the answer has settled, to read the provider's own Markdown
+# through the clipboard. Looked up inside the answer node, so it must be
+# relative to it. Leave empty to always convert the DOM instead.
+copy_button = "button[data-testid='copy-turn-action-button'], button[aria-label='Copy'], button[aria-label='Copier']"
+login_page_indicator = "button[data-testid='login'], form[action*='login'], a[href*='auth/login'], a[href*='auth/0']"
+captcha_indicator = "iframe[src*='captcha'], iframe[src*='challenges.cloudflare.com'], div#challenge-form"
+error_banner = "[data-testid='error-message'], div[role='alert']"
+
+[login]
+url_patterns = ["/auth/login", "/login", "accounts.google.com", "/auth/0"]
+# The page a signed-out visitor lands on (title "ChatGPT: Chat, Work, Create &
+# Code with AI") has no prompt field and shows these call to action buttons.
+# The text patterns are only used when that field is missing, so a signed-in
+# conversation is never mistaken for a sign-in screen.
+text_patterns = ["log in", "sign up", "create an account"]
+
+[rate_limit]
+text_patterns = [
+  "too many requests",
+  "rate limit",
+  "trop de requêtes",
+  "réessayez plus tard",
+  "you've reached",
+]
+
+[timeouts]
+page_load = 30
+response_generation = 120
+response_timeout = 180
+
+[input]
+poll_interval_ms = 400
+quiet_ms = 1500
+inject_method = "auto"
+
+[options]
+supports_images = true
+supports_streaming = true
+supports_new_chat = true
+
+# Multi-account. A ChatGPT account is a browser profile, so declaring two
+# accounts gives two independent sessions, each with its own tab pool, driven
+# side by side. Address them with the model string:
+#
+#   {"model": "chatgpt"}                              -> the default account
+#   {"model": "chatgpt@perso"}                        -> by account id
+#   {"model": "[email protected]@gmail.com"}     -> by address
+#   {"model": "chatgpt@perso/gpt-4o"}                 -> account and model
+#
+# Sign each one in once:  llm-gateway login chatgpt --account perso
+# Their profiles live in profiles/chatgpt/accounts/<account>/user-data-dir.
+#
+# [[accounts]]
+# id = "perso"
+# email = "[email protected]"
+# label = "Personal"
+# default = true
+#
+# [[accounts]]
+# id = "pro"
+# email = "[email protected]"
+# label = "Work"

providers/claude.toml

@@ -0,0 +1,79 @@
+# Claude (claude.ai) provider configuration, driven entirely by this file.
+# Selectors change when the web UI is updated: edit them here (reloaded live),
+# never in the Rust source. Validate with: llm-gateway selftest claude
+#
+# The values below come from the public claude.ai DOM observed in 2026 and from
+# published selector lists (May 2026). They are a starting point, not a
+# guarantee: run 'llm-gateway selftest claude' before relying on them.
+
+[provider]
+name = "claude"
+display_name = "Anthropic Claude"
+web_url = "https://claude.ai/new"
+new_conversation_url = "https://claude.ai/new"
+# Captures the web conversation id from the current URL (one capture group).
+conversation_url_pattern = '^https://claude\.ai/chat/([0-9a-zA-Z-]{8,})'
+default_models = ["claude-sonnet-4", "claude-opus-4", "claude-haiku-4"]
+enabled = true
+
+[selectors]
+# Claude uses ProseMirror, so the editor is a contenteditable div rather than a
+# textarea. The fallbacks cover the composer variants seen behind A/B tests.
+input_field = "div.ProseMirror[contenteditable='true'], div[contenteditable='true'][data-placeholder], [data-testid='composer-input']"
+send_button = "button[aria-label='Send Message'], button[aria-label='Send message'], button[data-testid='send-button'], button[type='submit']"
+# One node per assistant answer, in document order.
+response_container = "div.font-claude-response, div[data-is-streaming], div.font-claude-message"
+streaming_indicator = "button[aria-label='Stop response'], button[aria-label='Stop Response']"
+file_input = "input[type='file']"
+new_chat_button = "a[href='/new'], button[aria-label='New chat']"
+login_page_indicator = "a[href*='/login'], button[data-testid='login-button'], form[action*='login']"
+captcha_indicator = "iframe[src*='challenges.cloudflare.com'], iframe[title*='challenge'], div#challenge-form"
+error_banner = "[role='alert'], div[data-testid='error']"
+# Relative to the answer node: the hover toolbar button that copies the answer
+# as Markdown.
+copy_button = "button[aria-label='Copy'], button[data-testid='copy-button']"
+
+[login]
+url_patterns = ["/login", "/magic-link", "accounts.google.com"]
+# Only consulted when the prompt field is missing, so a chat page that happens
+# to mention "log in" is never misclassified.
+text_patterns = ["sign in", "log in", "continue with google"]
+
+[rate_limit]
+text_patterns = [
+  "message limit",
+  "rate limit",
+  "too many requests",
+  "you've reached",
+  "limite de messages",
+  "trop de requêtes",
+  "réessayez plus tard",
+]
+
+[timeouts]
+page_load = 30
+response_generation = 120
+response_timeout = 180
+
+[input]
+poll_interval_ms = 400
+quiet_ms = 1500
+inject_method = "auto"
+
+[options]
+supports_images = true
+supports_streaming = true
+supports_new_chat = true
+
+# Multi-account: each account is a separate browser profile, so several
+# subscriptions can be driven side by side. Declare them like this and address
+# them with model = "claude@work".
+#
+# [[accounts]]
+# id = "perso"
+# email = "[email protected]"
+# default = true
+#
+# [[accounts]]
+# id = "work"
+# email = "[email protected]"

providers/deepseek.toml

@@ -0,0 +1,73 @@
+# DeepSeek (chat.deepseek.com) provider configuration, driven entirely by this
+# file. Selectors change when the web UI is updated: edit them here (reloaded
+# live), never in the Rust source. Validate with: llm-gateway selftest deepseek
+#
+# The values below combine the public chat.deepseek.com DOM with published
+# selector lists (May 2026). They are a starting point, not a guarantee: run
+# 'llm-gateway selftest deepseek' before relying on them.
+
+[provider]
+name = "deepseek"
+display_name = "DeepSeek"
+web_url = "https://chat.deepseek.com/"
+new_conversation_url = "https://chat.deepseek.com/"
+# Captures the web conversation id from the current URL (one capture group).
+conversation_url_pattern = '^https://chat\.deepseek\.com/a/chat/s/([0-9a-zA-Z-]{8,})'
+default_models = ["deepseek-chat", "deepseek-reasoner"]
+enabled = true
+
+[selectors]
+# The composer is a plain <textarea>. DeepSeek localises its placeholder and the
+# id seen in older builds is gone, so the wording observed on the live page comes
+# first and the older guesses are kept as fallbacks.
+input_field = "textarea#chat-input, textarea[placeholder*='Message DeepSeek'], textarea[placeholder*='Envoyer'], textarea[placeholder*='Send a message'], textarea[placeholder*='message']"
+send_button = "div[role='button'][aria-label='Send'], button[aria-label='Send message'], button[data-testid='send-button']"
+# The answer body is rendered by the "ds-markdown" component; the attribute
+# first variant mirrors the ChatGPT markup.
+response_container = "div[data-message-author-role='assistant'], div.ds-markdown, div[class*='markdown-body']"
+streaming_indicator = "div[role='button'][aria-label*='Stop'], button[aria-label*='Stop'], div[class*='_stop']"
+file_input = "input[type='file']"
+new_chat_button = "a[href='/'], div[class*='new-chat'], button[aria-label='New chat']"
+login_page_indicator = "div[class*='ds-modal'], a[href*='sign_in'], button[data-testid='login-button']"
+captcha_indicator = "iframe[src*='captcha'], div[class*='captcha'], div#challenge-form"
+error_banner = "[role='alert'], div[class*='ds-toast'], div[class*='error']"
+# Relative to the answer node: the toolbar button that copies the answer.
+copy_button = "[aria-label='Copy'], [data-testid='copy-button'], div[class*='ds-icon-button'][role='button']"
+
+[login]
+url_patterns = ["/sign_in", "/login", "account.deepseek.com"]
+text_patterns = ["log in", "sign in", "log in with google", "se connecter"]
+
+[rate_limit]
+text_patterns = [
+  "too many requests",
+  "rate limit",
+  "please try again later",
+  "请求过于频繁",
+  "服务器繁忙",
+  "trop de requêtes",
+]
+
+[timeouts]
+page_load = 30
+response_generation = 120
+# DeepSeek R1 reasons for a long time before answering: keep a generous budget.
+response_timeout = 300
+
+[input]
+poll_interval_ms = 400
+quiet_ms = 1500
+inject_method = "auto"
+
+[options]
+supports_images = true
+supports_streaming = true
+supports_new_chat = true
+
+# Multi-account: each account is a separate browser profile. Address them with
+# model = "deepseek@perso".
+#
+# [[accounts]]
+# id = "perso"
+# email = "[email protected]"
+# default = true

providers/selectors.md

@@ -0,0 +1,169 @@
+# Provider selectors
+
+Every selector lives in a provider configuration file
+(`~/.llm-gateway/providers/<provider>.toml`), never in the Rust source. When a
+web UI changes, edit the TOML: the server picks the change up within a second
+(no restart, no browser relaunch). `llm-gateway selftest <provider>` tells you
+whether the current selectors still work.
+
+## Why this file exists
+
+Web UIs change without notice. The three failure modes we care about are:
+
+| Symptom | Reported as | What to do |
+|---|---|---|
+| The prompt field disappeared | HTTP 500, code `selector_missing` | Update `selectors.input_field` |
+| The send button disappeared | HTTP 500, code `selector_missing` | Update `selectors.send_button` |
+| The answer container moved | empty answer or HTTP 500 | Update `selectors.response_container` |
+| The streaming indicator is wrong | answers truncated or 504 | Update `selectors.streaming_indicator` |
+| The session expired | HTTP 401, code `requires_login` | Run `llm-gateway login <provider>` |
+
+A failing capture always leaves a screenshot and a DOM dump in
+`~/.llm-gateway/debug/`, plus one line per failure in `debug/index.jsonl`.
+
+## How to update a selector
+
+1. Open the provider page in a normal browser, signed in.
+2. Open the developer tools, inspect the element you need.
+3. Prefer stable hooks in this order: `data-testid` > `id` > semantic attributes
+   (`aria-label`, `role`, `type`) > a CSS class. Avoid generated class names such
+   as `.css-1x2y3z`: they change on every deploy.
+4. Put a comma separated list in the TOML when a UI has several variants
+   (`div#prompt-textarea, div.ProseMirror[contenteditable='true']`): the first
+   match wins and the fallbacks keep you running through a partial rollout.
+5. Run `llm-gateway selftest <provider>` and keep the TOML only when it reports
+   `ok`.
+
+## Required and optional selectors
+
+| Key | Required | Used for |
+|---|---|---|
+| `input_field` | yes | focusing the editor, verifying the injected text, detecting an empty editor after submit |
+| `send_button` | yes | submitting the prompt (Enter is the fallback) |
+| `response_container` | yes | one node per assistant answer, in document order; the driver reads the n-th node. It must select the **answer text**, not the whole exchange: a container that also holds the action toolbar would capture the word "Copy" as part of the answer |
+| `copy_button` | no | the per-answer "copy" control, which gives the answer as Markdown through the clipboard. It may be inside the answer node or a sibling of it: the driver looks for it there and never elsewhere on the page, so the copy button of a previous answer is never clicked |
+| `streaming_indicator` | recommended | element that only exists while the answer is generated (a stop button) |
+| `file_input` | for images | hidden input fed through `DOM.setFileInputFiles` |
+| `attachment_indicator` | optional | chip shown once an upload has been ingested |
+| `new_chat_button` | optional | starting a brand new conversation without reloading |
+| `login_page_indicator` | recommended | detecting a sign-in page even when the URL does not say so. On chatgpt.com the signed-out landing page exposes `button[data-testid='login']` |
+| `[login] text_patterns` | recommended | lowercase substrings of the visible text that reveal a signed-out page; only consulted when the prompt field is missing, so a chat page that mentions "log in" is never misclassified |
+| `captcha_indicator` | recommended | detecting a bot challenge |
+| `error_banner` | recommended | surfacing the upstream error message to the API client |
+| `model_selector` | reserved | not driven yet: the gateway never changes the model in the UI |
+
+## Answer fidelity (Markdown)
+
+A web UI renders its answer as rich HTML and `innerText` flattens it: a table
+collapses into loose lines and a code block loses its fences. Once the answer has
+settled the driver therefore reads it a second time, in this order:
+
+1. **the clipboard** — it clicks the provider's own copy button and reads the
+   clipboard over CDP (`Browser.setPermission` then
+   `navigator.clipboard.readText()`). The provider copies exactly the Markdown it
+   rendered, so nothing is guessed. Requires `selectors.copy_button`.
+2. **the DOM** — it converts the answer node's `innerHTML` to Markdown
+   (headings, lists, quotes, links, images, fenced code, GitHub tables).
+3. **the visible text** — the historical behaviour, used when both fail.
+
+`[capture] markdown` chooses between them and a provider may override it in its
+own `[input]` section. Streamed deltas always come from the DOM, so they stay
+plain text; the final answer — the one a non-streaming client receives — is the
+converted one, and the source used is logged as `fidelity`.
+
+## Current status
+
+| Provider | File | Last validated | Status |
+|---|---|---|---|
+| OpenAI ChatGPT | `providers/chatgpt.toml` | 2026-09-18, signed-in session | `selftest` ok; `fidelity = "clipboard"` |
+| Anthropic Claude | `providers/claude.toml` | 2026-09-18, signed-in session | `selftest` ok; `fidelity = "clipboard"` |
+| DeepSeek | `providers/deepseek.toml` | 2026-09-18, signed-in session | `selftest` ok; **`fidelity = "dom"`**: the copy button has not been observed yet |
+| Local fixture (tests only) | generated by `tests/e2e_browser.rs` | 2026-09-18 | validated: full pipeline, streaming, login, captcha, rate limit, error banner, timeout, missing selector, rewrite, attachment, Markdown fidelity |
+
+The Claude and DeepSeek selectors were collected from public sources rather than
+observed on a signed-in session. Treat them as a starting point: the first run of
+`llm-gateway selftest <provider>` tells you whether they still match, and the
+failure modes above tell you which key to fix.
+
+The ChatGPT selectors shipped in this repository are the ones observed on the
+public web UI in September 2026. They are a starting point, not a guarantee:
+run the selftest before relying on them.
+
+## Upgrading the installed files
+
+The provider files under `~/.llm-gateway/providers/` are **copies** made on the
+first run: editing the ones in this repository does not change them. After
+upgrading the binary, refresh them with:
+
+```bash
+llm-gateway config init --force     # keeps the previous file as <provider>.toml.bak
+```
+
+## Localised user interfaces
+
+A provider translates its own UI, and the browser locale of the machine does not
+change that: the same ChatGPT account serves a French page under `lang="fr-CA"`
+and an English one under `lang="en-US"`. Every `aria-label` fallback therefore
+only matches one language, which is why the stable hooks come first:
+
+| Observed (French UI, September 2026) | Selector to prefer |
+|---|---|
+| `<button type="submit" id="composer-submit-button" aria-label="Envoyer la requête" data-testid="send-button">` | `button[data-testid='send-button']`, `button#composer-submit-button` |
+| the answer toolbar copy control | `button[data-testid='copy-turn-action-button']` |
+| `<textarea placeholder="Message DeepSeek" name="search" rows="2">` | `textarea[placeholder*='Message DeepSeek']` - the English placeholder and the `#chat-input` id of older builds are gone |
+
+The `[login] text_patterns` follow the same rule: they are lowercase substrings of
+the page text, so a French sign-in page needs French patterns
+(`"se connecter"`, `"s'inscrire"`) next to the English ones. A page that is not
+recognised as a sign-in page is reported as a missing prompt field instead, which
+is a confusing way to learn that the session expired.
+
+## Several accounts of one provider
+
+A signed-in account is a browser profile, so driving two ChatGPT accounts means
+two profiles, two sessions and two tab pools. Declare them in the provider file:
+
+```toml
+[[accounts]]
+id = "perso"                          # the short name used in the model string
+email = "[email protected]"     # also accepted as the account key
+label = "Personal"                    # free-form, shown by 'llm-gateway list'
+default = true                        # used by {"model": "chatgpt"}
+
+[[accounts]]
+id = "pro"
+email = "[email protected]"
+```
+
+Then sign each one in once and address it in the model string:
+
+```bash
+llm-gateway login chatgpt --account perso
+llm-gateway login chatgpt@pro                  # the same thing
+```
+
+```jsonc
+{"model": "chatgpt"}                                  // the default account
+{"model": "chatgpt@pro"}                              // by id
+{"model": "[email protected]@gmail.com"}         // by address
+{"model": "chatgpt@pro/gpt-4o"}                       // account and model
+```
+
+Their profiles live in `profiles/<provider>/accounts/<account>/user-data-dir`.
+A provider that declares no account keeps the historical
+`profiles/<provider>/user-data-dir`, so adding accounts later means signing in
+again for the first one. Each account has its own tab pool, so `max_tabs`
+applies per account.
+
+## Adding a new provider
+
+1. Copy `providers/chatgpt.toml` to `~/.llm-gateway/providers/<name>.toml`.
+2. Set `provider.name`, `provider.web_url`, `provider.new_conversation_url` and
+   `provider.conversation_url_pattern` (one capture group, the web conversation
+   id extracted from the page URL).
+3. Fill in the selectors, then `llm-gateway selftest <name>`.
+4. `llm-gateway login <name>` once, and the provider is usable as a model:
+   `{"model": "<name>"}` or `{"model": "<name>/<model>"}`.
+
+Nothing else is provider specific: the capture loop, the error mapping, the
+conversation store and the API are shared.