You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
External integrations can now declare type "ai" and serve as the AI
backend of the assistant instead of Gladys Plus. The contract is fully
provider-agnostic: the whole agentic loop (MCP tool calling, prompts,
history) stays in the core, and the integration only receives
OpenAI-compatible chat completion requests over the integration
WebSocket (external-integration.ai.chat, 120s ack) and answers with
OpenAI-compatible completions — a thin adapter to Claude, DeepSeek,
OpenAI, a local LLM or any other provider.
Server:
manifest type "ai" (schema + validation), no device screens
ai.chat proxy service capability relayed by the supervisor
AI_PROVIDER system variable: gateway.aiChat routes all AI traffic
(chat loop, intent router, weekly digest) to the selected provider,
with the Gladys Plus model field stripped and no silent fallback
AI works without any Gladys Plus account (message.create)
GET/POST /api/v1/ai_provider endpoints, selection cleared on
uninstall of the selected provider
Front:
AI provider selection card on the AI integration page (admin)
generic external integration page branches for the ai type
install warning dedicated to AI providers, catalog placement
chat model selector hidden when an external provider is active
en/fr/de translations
Spec: docs/specs/external-integrations.md B.18 + C.1/C.4/C.5/C.8
updated in the same diff (spec-first rule).
External integrations can now declare type "ai" and serve as the AI
backend of the assistant instead of Gladys Plus. The contract is fully
provider-agnostic: the whole agentic loop (MCP tool calling, prompts,
history) stays in the core, and the integration only receives
OpenAI-compatible chat completion requests over the integration
WebSocket (external-integration.ai.chat, 120s ack) and answers with
OpenAI-compatible completions — a thin adapter to Claude, DeepSeek,
OpenAI, a local LLM or any other provider.
Server:
- manifest type "ai" (schema + validation), no device screens
- ai.chat proxy service capability relayed by the supervisor
- AI_PROVIDER system variable: gateway.aiChat routes all AI traffic
(chat loop, intent router, weekly digest) to the selected provider,
with the Gladys Plus model field stripped and no silent fallback
- AI works without any Gladys Plus account (message.create)
- GET/POST /api/v1/ai_provider endpoints, selection cleared on
uninstall of the selected provider
Front:
- AI provider selection card on the AI integration page (admin)
- generic external integration page branches for the ai type
- install warning dedicated to AI providers, catalog placement
- chat model selector hidden when an external provider is active
- en/fr/de translations
Spec: docs/specs/external-integrations.md B.18 + C.1/C.4/C.5/C.8
updated in the same diff (spec-first rule).
Community request: https://community.gladysassistant.com/t/10421
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.
Use the following commands to manage reviews:
@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.
Use the checkboxes below for quick actions:
▶️ Resume reviews
🔍 Trigger review
📝 Walkthrough
Walkthrough
This change adds ai external integrations with OpenAI-compatible chat over WebSocket. Administrators can select providers through new APIs and UI settings. AI requests route to the selected provider, with lifecycle cleanup, validation, translations, and tests.
Changes
External AI provider
Layer / File(s)
Summary
AI integration contract and proxy runtime docs/specs/external-integrations.md, server/lib/external-integration/*, server/utils/constants.js, server/test/lib/external-integration/*
AI manifests support type: "ai", the ai.chat capability, OpenAI-compatible payloads, and a 120-second timeout.
Provider selection and lifecycle server/api/controllers/externalIntegration.controller.js, server/api/routes.js, server/lib/external-integration/*, server/test/controllers/externalIntegration/*
New APIs list, select, and reset providers. Uninstalling the selected provider clears AI_PROVIDER.
Gateway and message routing server/lib/gateway/gateway.aiChat.js, server/lib/message/message.create.js, server/test/lib/gateway/*, server/test/lib/message/*
Configured providers receive AI requests without the Gladys Plus model or fallback. AI messages work without Gladys Plus configuration.
Provider configuration and integration UX front/src/routes/integration/*, front/src/routes/chat/AiModelSelector.jsx, front/src/config/i18n/*
Administrators can select providers. AI integrations route to configuration screens, hide device screens, suppress the model selector, and display localized warnings and error states.
A rabbit selects a provider bright,
Chat hops through WebSocket light.
The model hides when providers speak,
Admin settings make the choice unique.
When one is removed, its setting clears.
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Review of Add generic AI provider external integration type (#2744).
Solid, spec-first design: provider-agnostic OpenAI-compatible relay, explicit admin selection, no silent Gladys Plus fallback, install trust warning, and AI without Plus via message.create. Tests cover the new relay, routing, selection, uninstall cleanup, and API RBAC. No brand-specific device categories in constants.js.
Not approving. This is the highest-trust external-integration class (conversations + tool results + device/scene authority). Labeled risk:high and needs:human-review; requesting @atrovato (author is already Pierre-Gilles).
Must resolve before merge
Spec section number collision with #2738 — that open PR already claims B.18 for weather. This branch should use B.19 (or renumber weather), including the phase-2 table and C.1/C.4/C.5/C.8 cross-refs.
Living-spec banner still outdated — the document header still says “B.15–B.17 are phase 2 designs not yet implemented” while this PR ships a new B.18. Update the banner in the same diff (weather’s branch already rewrote it).
Incomplete ai front redirects — ExternalIntegrationPage correctly drops device tabs for ai, and install/catalog URLs go to /config, but device-page/index.js and discover-page/index.js still only redirect communication. Direct /device or /discover URLs for an AI provider can still hit empty device screens.
Also note
Vendored manifest.schema.json accepts ai; the store indexer (GladysAssistant/integration-store) and SDK (onAiChat) need matching companion updates or catalog installs/onAiChat will lag the core.
Completion validation only rejects non-objects; empty {} still reaches extractAssistantMessage and fails softer as openai.request.fail. Tightening is optional.
The reason will be displayed to describe this comment to others. Learn more.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@server/lib/external-integration/externalIntegration.aiChat.js`:
- Around line 26-29: Strengthen the completion validation in the result-handling
code by requiring a non-empty choices array and an object at choices[0].message
before returning the response. Preserve throwing
ExternalIntegrationUnavailableError with
EXTERNAL_INTEGRATION_INVALID_AI_RESPONSE for all malformed structures, and add
unit-test cases for an empty object, an empty choices array, and a missing or
invalid first message.
In `@server/lib/external-integration/externalIntegration.uninstall.js`:
- Around line 55-60: Serialize setAiProvider() with uninstall() for the same
integration so provider selection cannot be written after uninstall has checked
or removed the service. Update the synchronization or availability validation
around these methods, preserving normal provider selection while ensuring
AI_PROVIDER is never saved with the uninstalling service.selector.
In `@server/test/lib/message/message.create.test.js`:
- Around line 76-105: The test around MessageHandler.create must also verify
that the Gladys Plus-required fallback event is not emitted when an external AI
provider is configured. Extend the existing event.emit assertions in the test to
assert no openai.plus-required reply is sent, while preserving the current
message.new-for-open-ai assertion and covering the new branch.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
Push a commit to this branch (recommended)
Create a new PR with the fixes
ℹ️ Review info⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: c7907e60-adea-4387-8623-4c289a9f7fa4
📥 Commits
Reviewing files that changed from the base of the PR and between c4e452a and 36eaa96.
- Renumber spec section B.18 to B.19 (B.18 is taken by the weather type
in the parallel PR #2738) and update the living-spec banner to reflect
the implemented state of B.15-B.17 and B.19
- Redirect direct URL access to the device/discover screens of an AI
integration to its configuration screen (same rule as communication)
- Validate the ai.chat completion at the contract boundary: data must
carry an object choices[0].message, so adapter bugs surface as an
explicit invalid-response error instead of a silent empty turn
- Guard setAiProvider against a concurrent uninstall of the selected
provider (re-check after write, roll back the variable)
- Assert the plus-required reply is not sent when an external AI
provider serves the assistant (message.create test)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
The reason will be displayed to describe this comment to others. Learn more.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@server/lib/external-integration/externalIntegration.setAiProvider.js`:
- Around line 27-37: Make the AI-provider selection and
externalIntegration.uninstall cleanup atomic rather than relying on the
post-write stillInstalled check in setAiProvider. Serialize both operations or
use an atomic conditional update/delete so uninstall cannot leave or recreate
AI_PROVIDER with the removed service_id; add a concurrency test covering the
interleaving.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
Push a commit to this branch (recommended)
Create a new PR with the fixes
ℹ️ Review info⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: b64b00b9-b31b-45d3-9a69-29406ef2950d
📥 Commits
Reviewing files that changed from the base of the PR and between 36eaa96 and f6898f6.
…ndbox fix
Review fixes (cursor bot, blocking):
- renumber the TTS provider section from B.18 to B.20 everywhere (spec,
schema comments, code comments): B.18 is owned by the weather type
(#2738) and B.19 by the AI provider type (#2744)
- clear TTS_ACTIVE_PROVIDER when uninstalling the selected provider
(the AI_PROVIDER pattern of #2744): no dangling selector failing every
announcement; documented in B.20 and C.5, covered by tests
- living-spec truthfulness: the banner now states B.15-B.17 are
implemented and reserves B.18/B.19 for the parallel PRs; the
"current state" block of B.20 is rewritten as pre-refactor state
Review fixes (cursor bot + CodeRabbit, non-blocking):
- GET /api/v1/tts/provider now returns a display name per provider (the
integration manifest name) so the System settings select matches the
Integrations UI instead of showing raw selectors
- getLocalApiBaseUrl prefers RFC1918 IPv4 addresses on multi-homed hosts
(VPN tunnels or virtual bridges enumerating first)
- own-key lookup on the TTS content-type allow-list (inherited
Object.prototype keys such as "constructor" are rejected)
- defense-in-depth validation in tts.getSpeechUrl: buffer type, size
bound and curated content type/extension are re-checked before any
byte is cached or served, whatever service implements the provider
interface
- TTS provider settings card commits the selection only after the POST
succeeds (select disabled while saving, reload on error)
- device/discover pages resolve the integration type before firing
device requests, so config-only integrations redirect cleanly
- the tts catalog category is hidden from non-admin users
- mirrored unit tests for tts.getProviderConfiguration and
tts.purgeExpiredAudios
CI fix (Server test failure, 300+ cascading failures):
- tts.getLocalApiBaseUrl.test.js called sinon.restore() on the DEFAULT
sandbox in afterEach, untracking the fakes other test files create at
load time: their sinon.reset() cleanup stopped clearing call history
and every suite running after the file leaked call counts ("called
twice" assertions, fake-timer timeouts). The stub now lives in a
dedicated sinon sandbox (the documented test/models/index.test.js
pattern). Full suite verified locally: only the environment-specific
failures of this sandbox remain (outbound proxy vs nock, missing
sqlite3 binary).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013ihJhVk52zW6qUsxBHeqx2
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Re-review of f6898f6 (follow-up to the first Cursor pass).
Verdict
Architecture matches B.19 well: provider-agnostic OpenAI-compatible relay, agentic loop stays in core, explicit admin selection, no silent Gladys Plus fallback, uninstall clears AI_PROVIDER, AI stays admin-managed, and non-admin list/detail RBAC still filters to communication only. Previous blockers (B.18→B.19 renumber, living-spec banner, device/discover redirects, completion choices[0].message validation, concurrent-uninstall re-check) are addressed. CI is green including Codecov patch.
Not approving: this is the highest-trust external-integration class (conversations + tool results + home/scene authority). Keeping risk:high + needs:human-review for a maintainer product pass. @atrovato is already requested.
Remaining notes
Trust warning coverage — the store install page warns for type: "ai", but until the indexer accepts ai, GitHub/dev install is the practical path and shows no equivalent warning. Selection in AiProviderSettings is immediate with no restated trust copy either (see inline).
AI page upsell — with Plus disconnected, the Gladys Plus upsell still dominates the AI page even when an external provider is selected and chat already works via message.create (see inline).
Soft residuals (ok for human review, not blocking): instance-wide single provider for chat/intent/digest; companion indexer/SDK onAiChat still needed for catalog listing (graceful until then per Pierre’s note).
No new brand-tied device categories/types in server/utils/constants.js — only AI_PROVIDER + external-integration.ai.chat.
…us upsell
Cursor re-review follow-ups:
- The trust warning shown on the store install screen never appears on
the GitHub/dev install path, and selection is the moment conversations
actually start flowing to the provider: the AI provider settings card
now restates the warning whenever an external provider is selected
- With an external provider active, the assistant already works without
Gladys Plus: the Plus upsell and the Plus rate-limit note are hidden
on the AI page in that case
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
Update externalProviderActive after provider selection.
getAiProvider() runs only in componentDidMount(). AiProviderSettings.updateProvider() updates the child state after POST /api/v1/ai_provider, but OpenAIGateway receives no update. Selecting an external provider therefore does not hide the upsell or rate-limit message on the current page. Resetting to Gladys Plus can leave those messages hidden until reload.
Pass an onProviderChange callback to AiProviderSettings, or lift the selector state into OpenAIGateway, and update it after the save succeeds.
The supplied front/src/routes/integration/all/openai/AiProviderSettings.jsx:13-52 snippet shows that the child updates only its own state.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@front/src/routes/integration/all/openai/index.js` at line 141, Update
OpenAIGateway’s provider-selection flow so externalProviderActive is refreshed
after AiProviderSettings.updateProvider() successfully saves a provider. Pass an
onProviderChange callback to AiProviderSettings or lift the state into
OpenAIGateway, ensuring both external-provider selection and resetting to Gladys
Plus immediately update the current page’s upsell and rate-limit messaging.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@front/src/routes/integration/all/openai/index.js`:
- Around line 30-39: Update getAiProvider and the externalProviderActive
initialization to use null as the unknown state, preserving false only when the
provider response confirms no external provider. In the affected Gladys Plus
upsell and rate-limit render conditions, require externalProviderActive ===
false so messages remain hidden while the request is pending or fails, while
retaining the external-provider-active behavior.
---
Outside diff comments:
In `@front/src/routes/integration/all/openai/index.js`:
- Line 141: Update OpenAIGateway’s provider-selection flow so
externalProviderActive is refreshed after AiProviderSettings.updateProvider()
successfully saves a provider. Pass an onProviderChange callback to
AiProviderSettings or lift the state into OpenAIGateway, ensuring both
external-provider selection and resetting to Gladys Plus immediately update the
current page’s upsell and rate-limit messaging.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
Push a commit to this branch (recommended)
Create a new PR with the fixes
ℹ️ Review info⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: eb8c09ac-7916-41e7-8be4-33710f1a231c
📥 Commits
Reviewing files that changed from the base of the PR and between f6898f6 and ddd7a80.
CodeRabbit follow-ups on the AI page:
- externalProviderActive starts as null (unknown): the Plus upsell and
rate-limit messages only render once the absence of an external
provider is confirmed, instead of flashing while the request is
pending or lying after a failed request
- AiProviderSettings notifies the page through onProviderChange, so
selecting or resetting the provider updates the messaging without a
reload
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
The reason will be displayed to describe this comment to others. Learn more.
Re-review of bdbd983 (follow-up to the Cursor pass on f6898f6).
Verdict
Architecture still matches B.19 well: provider-agnostic OpenAI-compatible relay, agentic loop stays in core, explicit admin selection, no silent Gladys Plus fallback, uninstall clears AI_PROVIDER, AI stays admin-managed, non-admin list/detail RBAC still filters to communication only. Previous soft notes are fixed in ddd7a802 / bdbd983:
trust warning restated under the provider select
Plus upsell / rate-limit copy gated on confirmed absence of an external provider (null while loading/failed)
parent page kept in sync via onProviderChange
CI is green (including Codecov patch). No brand-tied device categories/types — only AI_PROVIDER + external-integration.ai.chat.
Not approving. This remains the highest-trust external-integration class (conversations + tool results + home/scene authority via the tool-calling loop). Keeping risk:high + needs:human-review for a maintainer product pass. @atrovato is already requested.
Soft residual for human review
Plus connected + external provider — when Gladys Plus is configured and an external provider is selected, the page still shows the Plus “chat enabled” success line, AiQuotaDisplay, and Plus-only debug download. Quota especially is misleading once chat traffic no longer hits Plus (see inline).
Companion store indexer / SDK onAiChat still needed for catalog listing (graceful until then; GitHub/dev install remains the practical path).
Living-spec body: B.15–B.17 section titles still say “(design, phase 2)” while the banner correctly treats them as shipped — cosmetic inconsistency across parallel provider PRs, not a merge blocker for this change.
…tive
With Gladys Plus connected and an external provider selected, the AI
traffic no longer hits Plus: the Plus quota display and the 'chat
enabled by default' banner were misleading in that state. Both are now
gated on the confirmed absence of an external provider, like the upsell
and rate-limit messages. The weekly digest and debug context download
stay: their LLM calls follow the selected provider through
gateway.aiChat.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
Keep the post-create refresh on the originating page.
Capture pageGeneration before the POST. If the selector changes while the POST is pending, return before calling getDiscoveredDevices(). Otherwise, that method uses the new selector and updates the wrong page. After unmount, it can also call setState before its own generation guard runs.
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@front/src/routes/integration/all/external-integration/discover-page/index.js`
around lines 133 - 149, Update createDevice to capture the current
pageGeneration before posting, then verify it is unchanged after the POST and
return if the selector changed or the component is no longer current. Only call
getDiscoveredDevices when the originating page generation remains valid,
preventing refreshes against a different page or after unmount.
Do not report success when the scan completion event is missing.
When the 60-second timer invokes finishScan, the method sets RequestStatus.Success but does not call getDiscoveredDevices(). The page can show success with the old device list. A later POST failure is also ignored after scanToken changes. Use an explicit timeout/error state, or perform an authoritative refresh before reporting success.
Also applies to: 110-120
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@front/src/routes/integration/all/external-integration/discover-page/index.js`
around lines 87 - 89, Update the scan timeout flow around finishScan so a
missing scan completion event cannot set RequestStatus.Success with stale
devices. On timeout, either set an explicit error/timeout status or fetch the
authoritative device list with getDiscoveredDevices() before reporting success;
ensure late POST failures after scanToken changes are not silently ignored.
Cancel the debounced search when the page loses focus.
loadIntegrationPage() and componentWillUnmount() bump pageGeneration and clear the scan timer, but a pending debounce can still run and call setState with the old selector. Use debounce(...).clear() at those same points. If searching is page-scoped, reset deviceSearch when the debounced call is cleared.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@front/src/routes/integration/all/external-integration/discover-page/index.js`
around lines 20 - 22, Update loadIntegrationPage() and componentWillUnmount() to
clear the pending debouncedSearchDevices invocation alongside the existing
pageGeneration and scan-timer cleanup, using its debounce clear API. When
clearing the page-scoped search, also reset deviceSearch so stale selector
results cannot update state after focus loss or unmount.
pageGeneration changes only when the selector changes. The websocket handler and createDevice can start several requests for one generation. If an older request resolves after a newer request, Lines 67 and 73 overwrite the current list or status with stale data. Track a request sequence token and check it in both the success and error paths.
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@front/src/routes/integration/all/external-integration/discover-page/index.js`
around lines 58 - 74, Update the discovered-device refresh flow around the
request in the component method containing pageGeneration to track a
monotonically increasing request sequence token for every request, including
websocket and createDevice-triggered refreshes within the same generation.
Capture the token before awaiting httpClient.get, and require both the
generation and token to still match before applying discoveredDevices or
GetDiscoveredDevicesStatus in the success and error paths.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In
`@front/src/routes/integration/all/external-integration/discover-page/index.js`:
- Around line 133-149: Update createDevice to capture the current pageGeneration
before posting, then verify it is unchanged after the POST and return if the
selector changed or the component is no longer current. Only call
getDiscoveredDevices when the originating page generation remains valid,
preventing refreshes against a different page or after unmount.
- Around line 87-89: Update the scan timeout flow around finishScan so a missing
scan completion event cannot set RequestStatus.Success with stale devices. On
timeout, either set an explicit error/timeout status or fetch the authoritative
device list with getDiscoveredDevices() before reporting success; ensure late
POST failures after scanToken changes are not silently ignored.
- Around line 20-22: Update loadIntegrationPage() and componentWillUnmount() to
clear the pending debouncedSearchDevices invocation alongside the existing
pageGeneration and scan-timer cleanup, using its debounce clear API. When
clearing the page-scoped search, also reset deviceSearch so stale selector
results cannot update state after focus loss or unmount.
- Around line 58-74: Update the discovered-device refresh flow around the
request in the component method containing pageGeneration to track a
monotonically increasing request sequence token for every request, including
websocket and createDevice-triggered refreshes within the same generation.
Capture the token before awaiting httpClient.get, and require both the
generation and token to still match before applying discoveredDevices or
GetDiscoveredDevicesStatus in the success and error paths.
ℹ️ Review info⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 779faed4-9e1c-4afc-8dbf-49f61342e7c5
📥 Commits
Reviewing files that changed from the base of the PR and between bdbd983 and 6d43ed9.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
area:aiAI agent, routing, promptsarea:frontPreact front-endarea:serverNode.js server codeneeds:human-reviewAutomated review is not confident, maintainer must take a lookrisk:highTouches DB migrations, auth, billing or user data. Careful human review requiredtype:featureNew user-facing feature or improvement
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
github-actions
Bot
added
area:server
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Pierre-Gilles
added
needs:human-review
Cursor
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
External integrations can now declare type "ai" and serve as the AI
backend of the assistant instead of Gladys Plus. The contract is fully
provider-agnostic: the whole agentic loop (MCP tool calling, prompts,
history) stays in the core, and the integration only receives
OpenAI-compatible chat completion requests over the integration
WebSocket (external-integration.ai.chat, 120s ack) and answers with
OpenAI-compatible completions — a thin adapter to Claude, DeepSeek,
OpenAI, a local LLM or any other provider.
Server:
(chat loop, intent router, weekly digest) to the selected provider,
with the Gladys Plus model field stripped and no silent fallback
uninstall of the selected provider
Front:
Spec: docs/specs/external-integrations.md B.18 + C.1/C.4/C.5/C.8
updated in the same diff (spec-first rule).
Community request: https://community.gladysassistant.com/t/10421
Co-Authored-By: Claude Fable 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01XhBhvoV4jdZEhxgadWwXV8
Summary by CodeRabbit