deploy(vera): back to the local lane — smart is stable again#17
Merged
Conversation
Reverts the 2026-07-23 cloud pin. The ML node stabilised and :8041 now serves 262144 tokens (verified against the live endpoint), roughly 7x the panel's ~37k-token finder prompts — so the failure that forced the pin is gone. That was a 32768-token TOTAL ceiling (prompt + output) which every large-diff review overran, falling back to Groq's free tier and its 12000 TPM limit, exhausting the panel and leaving PRs unreviewed. Deleting the block returns to the host default (protolabs/reasoning, aliased to protolabs/smart), which is where model settings belong — they are host-scoped by ADR 0047 and the agent-layer pin only existed to outrank the box default during the outage. Applied to the live config volume and restarted: LangGraph agent initialized (model: protolabs/reasoning). Local inference also takes the panel back to zero marginal cost per review. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Reverts the cloud pin.
The ML node stabilised and
:8041now serves 262,144 tokens — verified against the live endpoint, ~7× the panel's ~37k-token finder prompts. The failure that forced the pin is gone.For the record, that failure was: a 32,768-token TOTAL ceiling (prompt + output, confirmed empirically) which every large-diff review overran, falling back to Groq's free tier and its 12,000 TPM limit — the panel exhausted repeatedly and left PRs unreviewed with no verdict.
Deleting the block returns to the host default (
protolabs/reasoning→ aliased toprotolabs/smart), which is where model settings belong: they're host-scoped under ADR 0047, and the agent-layer pin existed only to outrank the box default during the outage.Already applied to the live config volume and restarted —
LangGraph agent initialized (model: protolabs/reasoning, ...), healthy. This PR keeps the seed in step so a fresh instance matches.Also takes the panel back to zero marginal cost per review (cloud was roughly $0.12–0.15 a review on DeepSeek-V4).
🤖 Generated with Claude Code