From 5706e3d2af0c4e8b3829d12643b90332ddcc89ee Mon Sep 17 00:00:00 2001
From: Kieran Klukas <kieran@dunkirk.sh>
Date: Tue, 16 Jun 2026 12:15:11 -0400
Subject: [PATCH] feat: prune old models

---
 server/db/queries/models_catalog.sql        |  5 +++++
 server/internal/pool/models_refresher.go    | 10 +++++++++-
 server/internal/store/models_catalog.sql.go | 21 +++++++++++++++++++++
 server/internal/store/querier.go            |  3 +++
 4 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/server/db/queries/models_catalog.sql b/server/db/queries/models_catalog.sql
index 0d782d6..c202e38 100644
--- a/server/db/queries/models_catalog.sql
+++ b/server/db/queries/models_catalog.sql
@@ -27,3 +27,8 @@ ORDER BY tier ASC, label ASC;
 -- name: GetModelCatalogRefreshedAt :one
 -- Oldest refreshed_at across all models — tells us when the catalog is stale.
 SELECT COALESCE(MIN(refreshed_at), 0) FROM models_catalog;
+
+-- name: PruneStaleModels :execrows
+-- Delete models for a provider prefix that weren't refreshed in the current cycle.
+-- The first arg is the LIKE pattern (e.g. "nvidia/%").
+DELETE FROM models_catalog WHERE id LIKE ? AND refreshed_at < ?;
diff --git a/server/internal/pool/models_refresher.go b/server/internal/pool/models_refresher.go
index 0f6c3b4..3dc45d4 100644
--- a/server/internal/pool/models_refresher.go
+++ b/server/internal/pool/models_refresher.go
@@ -70,6 +70,7 @@ func (r *ModelsRefresher) refreshAll(ctx context.Context) {
 		}
 
 		fetchCtx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+		now := time.Now().Unix()
 		models, err := fetcher.FetchModels(fetchCtx, r.httpClient, p.BaseURL, apiKey)
 		cancel()
 
@@ -85,7 +86,14 @@ func (r *ModelsRefresher) refreshAll(ctx context.Context) {
 				upserted++
 			}
 		}
+
+		// Prune models that no longer exist on this provider.
+		pruned, _ := r.q.PruneStaleModels(context.Background(), store.PruneStaleModelsParams{
+			ID:          p.ID + "/%",
+			RefreshedAt: now,
+		})
+
 		r.log.Info("models refresher: catalog updated",
-			"provider", p.ID, "models", len(models), "upserted", upserted)
+			"provider", p.ID, "models", len(models), "upserted", upserted, "pruned", pruned)
 	}
 }
diff --git a/server/internal/store/models_catalog.sql.go b/server/internal/store/models_catalog.sql.go
index ea8e497..b91fe9e 100644
--- a/server/internal/store/models_catalog.sql.go
+++ b/server/internal/store/models_catalog.sql.go
@@ -63,6 +63,27 @@ func (q *Queries) ListModelCatalog(ctx context.Context) ([]ModelsCatalog, error)
 	return items, nil
 }
 
+const pruneStaleModels = `-- name: PruneStaleModels :execrows
+g;
+
+DELETE FROM models_catalog WHERE id LIKE ? AND refreshed_at <
+`
+
+type PruneStaleModelsParams struct {
+	ID          string `json:"id"`
+	RefreshedAt int64  `json:"refreshed_at"`
+}
+
+// Delete models for a provider prefix that weren't refreshed in the current cycle.
+// The first arg is the LIKE pattern (e.g. "nvidia/%").
+func (q *Queries) PruneStaleModels(ctx context.Context, arg PruneStaleModelsParams) (int64, error) {
+	result, err := q.db.ExecContext(ctx, pruneStaleModels, arg.ID, arg.RefreshedAt)
+	if err != nil {
+		return 0, err
+	}
+	return result.RowsAffected()
+}
+
 const upsertModelCatalog = `-- name: UpsertModelCatalog :exec
 
 INSERT INTO models_catalog (
diff --git a/server/internal/store/querier.go b/server/internal/store/querier.go
index 31927cb..9b2d0c0 100644
--- a/server/internal/store/querier.go
+++ b/server/internal/store/querier.go
@@ -174,6 +174,9 @@ type Querier interface {
 	// Total pioneer credit limit across all active pool keys.
 	PoolTotalBalance(ctx context.Context) (interface{}, error)
 	PoolUserCount(ctx context.Context) (int64, error)
+	// Delete models for a provider prefix that weren't refreshed in the current cycle.
+	// The first arg is the LIKE pattern (e.g. "nvidia/%").
+	PruneStaleModels(ctx context.Context, arg PruneStaleModelsParams) (int64, error)
 	PutIdempotency(ctx context.Context, arg PutIdempotencyParams) error
 	// Called after a request settles. Resets daily counter if the day rolled over.
 	// today_day is the current UTC day (unix / 86400).
-- 
2.51.2