AI communication fails with online model despite chat test confirming it works #63

Closed
opened 2026-07-20 18:58:40 +00:00 by hugo · 1 comment
Owner

I set up an online and an offline inference provider in the preferences for my default project rfc1437. both work fine when fetching model lists and in the chat test. but when I use the online model to get for example a translation or taxonomy suggestions, I get an error about communication. When I switch to in-flight mode and use the offline provider (the localhost oMLX), it works fine.

I set up an online and an offline inference provider in the preferences for my default project rfc1437. both work fine when fetching model lists and in the chat test. but when I use the online model to get for example a translation or taxonomy suggestions, I get an error about communication. When I switch to in-flight mode and use the offline provider (the localhost oMLX), it works fine.
hugo added the bug label 2026-07-20 18:58:40 +00:00
Author
Owner

Implemented in 2f7d752 and pushed to origin/main. One-shot AI requests now use a provider-portable OpenAI-compatible Chat Completions payload without the OpenAI-specific strict response_format, restore the 16,384-token output budget, separate the 5-second connection timeout from the 120-second response timeout, and decode bare, fenced, or prose-wrapped JSON objects as bDS2 does. The shared path covers editor AI operations, automatic translation, import taxonomy mapping, gallery enrichment, and scripting. Verification: targeted AI tests (22 passed), cargo clippy -p bds-core --all-targets -- -D warnings, cargo build --workspace, cargo test --workspace (bds-core 480 passed/1 ignored; bds-ui 169 passed; remaining workspace and doc tests passed), git diff --check, allium check specs, and allium analyse specs. A neutral comparison against issue #63, bDS2, and specs/ai.allium found no remaining implementation gap.

Implemented in 2f7d752 and pushed to origin/main. One-shot AI requests now use a provider-portable OpenAI-compatible Chat Completions payload without the OpenAI-specific strict response_format, restore the 16,384-token output budget, separate the 5-second connection timeout from the 120-second response timeout, and decode bare, fenced, or prose-wrapped JSON objects as bDS2 does. The shared path covers editor AI operations, automatic translation, import taxonomy mapping, gallery enrichment, and scripting. Verification: targeted AI tests (22 passed), cargo clippy -p bds-core --all-targets -- -D warnings, cargo build --workspace, cargo test --workspace (bds-core 480 passed/1 ignored; bds-ui 169 passed; remaining workspace and doc tests passed), git diff --check, allium check specs, and allium analyse specs. A neutral comparison against issue #63, bDS2, and specs/ai.allium found no remaining implementation gap.
hugo closed this issue 2026-07-21 18:04:14 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: hugo/RuDS#63