deploy(vera): pin the cloud model while smart is pulled#13
Merged
Conversation
protoAgent's default model_name is `protolabs/reasoning`, which the gateway aliases to `protolabs/smart` — the local :8041 lane. With smart being pulled, an unpinned Vera goes down with it. `protolabs/cloud` -> deepseek-v4-pro (1M context). That also clears the failure that exhausted the panel this morning: the local lane served 32768 tokens TOTAL (prompt + output, verified empirically) against ~36.8k-token finder prompts, so large-diff reviews failed on context length after the Groq fallback failed on TPM. Model settings are host-scoped by default (ADR 0047); this agent-layer pin deliberately outranks the box default and should be deleted once smart returns. Applied to the LIVE config volume as well — the seed is copy-once and would otherwise only reach a fresh instance. Note: the gateway has no context profile for this alias, so compaction degrades from 'fraction:0.8' to 'messages:60'. Harmless at 1M context, worth knowing. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Keeps Vera alive while the local
smartbackend comes down.Why a pin is required, not optional
protoAgent's default
model_nameisprotolabs/reasoning, and the gateway aliases that toprotolabs/smart— the local :8041 lane. An unpinned Vera goes down with the backend.protolabs/cloud→ deepseek-v4-pro, 1M context.It also fixes this morning's exhaustion
The panel exhausted repeatedly on projectBoard-plugin#94/#95 — every step failing, no verdict posted, PRs left unreviewed. Two stacked causes:
"your prompt contains at least 32753 input tokens... for a total of at least 32769") against ~36.8k-token finder prompts.1M context clears both with room for the prior-round context to grow.
Scope
Model settings are host-scoped by default (ADR 0047); this agent-layer pin deliberately outranks the box default. Delete the block once smart is back.
Applied to the live config volume too — the seed is copy-once and would otherwise only reach a fresh instance. Live Vera is already running on it:
LangGraph agent initialized (model: protolabs/cloud, ...).Known degradation
The gateway has no context profile for this alias, so compaction falls back from
fraction:0.8tomessages:60. Harmless at 1M context, but it means compaction triggers on message count rather than actual context pressure.🤖 Generated with Claude Code