feat(knowledge): expose Cohere reranker controls

[waleedlatif1]

Summary

• Add self-hosted Cohere API key field on the knowledge block, mirroring the agent block's hosted-key pattern (getCohereRerankerApiKeyCondition) so BYOK works for self-hosted while staying invisible on hosted
• Add rerankerInputCount (1–100) advanced field to control how many vector results are sent to the Cohere reranker (defaults to 4× topK, capped at 100)
• Surface meta.warnings from Cohere /v2/rerank responses via logger.warn
• All new contract fields are optional + nullable — fully backwards compatible with existing hosted users

Test plan

• Hosted: reranker still works without an API key field appearing
• Self-hosted: Cohere API key field appears when reranker is enabled
• rerankerInputCount clamps to [1, 100] and respects the topK floor
• Cohere meta warnings show up in logs when present
• bun run check:api-validation passes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		May 4, 2026 6:11pm

[cursor]

PR Summary

Medium Risk
Moderate risk: changes touch search request validation and reranker API-key handling, which could affect rerank behavior/costs or accidentally expose/require credentials in self-hosted setups.

Overview
Adds new Cohere reranker controls to Knowledge search, including an advanced rerankerInputCount to tune how many vector candidates are sent for reranking (clamped 1–100 and never below topK).

Supports self-hosted Cohere usage by adding an optional per-request rerankerApiKey (surfaced as a conditional Knowledge block field) while keeping the field hidden on hosted or when NEXT_PUBLIC_COHERE_CONFIGURED is set; reranker key resolution now prefers user key on self-hosted, otherwise BYOK/env/rotation. Also logs Cohere /v2/rerank meta.warnings and documents the new env/Helm settings for Cohere configuration.

^{Reviewed by Cursor Bugbot for commit 5803229. Configure here.}

[greptile-apps]

Greptile Summary

This PR exposes three new Cohere reranker controls on the Knowledge block: a BYOK apiKey field (hidden on hosted and when NEXT_PUBLIC_COHERE_CONFIGURED=true), a rerankerInputCount advanced field to control oversample size (1–100, defaults to topK × 4), and meta.warnings surfacing from Cohere rerank responses. All new contract fields are optional/nullable and fully backwards-compatible.

Confidence Score: 5/5

Safe to merge — no bugs found; previous review feedback was addressed and all new fields are backwards-compatible.

All changed files reviewed thoroughly. The key resolution chain, field visibility conditions, Zod schema, client-side clamping, and server-side guard logic are all correct. The required: true re-introduction in commit 232b1ca is intentional and correctly paired with the isCohereConfigured hide-field mechanism. No P0 or P1 issues found.

No files require special attention.

Important Files Changed

Filename	Overview
apps/sim/lib/knowledge/reranker.ts	Adds userApiKey param to resolveCohereKey, adds meta.warnings surfacing via logger.warn, and extends CohereRerankResponse interface. Logic is correct: self-hosted key takes priority, BYOK and env fallbacks are preserved.
apps/sim/blocks/blocks/knowledge.ts	Adds rerankerInputCount (advanced short-input) and apiKey (Cohere key, hidden via getCohereRerankerApiKeyCondition). Both inputs registered in inputs schema. required: true on apiKey is intentional — field is hidden by the condition when isCohereConfigured.
apps/sim/blocks/utils.ts	Adds getCohereRerankerApiKeyCondition following the exact same pattern as getApiKeyCondition. Returns a thunk that hides the field on hosted or when isCohereConfigured, otherwise shows it when rerankerEnabled for search ops.
apps/sim/lib/api/contracts/knowledge/search.ts	Adds rerankerInputCount (integer 1–100, optional/nullable with null→undefined transform) and rerankerApiKey (optional string with empty→undefined transform). Both fields are backwards-compatible.
apps/sim/app/api/knowledge/search/route.ts	Honors rerankerInputCount for candidateTopK with topK floor and 100 ceiling; logs a warning when the value is raised; passes rerankerApiKey through to rerank(). Logic is correct.
apps/sim/tools/knowledge/search.ts	Client-side parsing of rerankerInputCount (string→integer, clamped 1–100) and apiKey; both conditionally spread into the request body only when reranking is enabled.
apps/sim/lib/core/config/feature-flags.ts	Adds isCohereConfigured flag from NEXT_PUBLIC_COHERE_CONFIGURED env var, following the same isTruthy(getEnv(...)) pattern as isAzureConfigured.
apps/sim/lib/core/config/env.ts	Adds NEXT_PUBLIC_COHERE_CONFIGURED to the createEnv public schema. Consistent with existing NEXT_PUBLIC_* optional string pattern.
apps/sim/.env.example	Documents both COHERE_API_KEY and NEXT_PUBLIC_COHERE_CONFIGURED, clearly pairing them as a required two-var setup for self-hosted deployments with pre-configured Cohere keys.
helm/sim/values.yaml	Adds NEXT_PUBLIC_COHERE_CONFIGURED with documentation comments, mirroring the existing Azure pattern.

_{Reviews (3): Last reviewed commit: "fix(knowledge): treat empty rerankerInpu..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 5803229. Configure here.}