feat(python): rollout store bindings (local + remote)

crates/lance-context-api/src/lib.rs

@@ -666,6 +666,8 @@ pub struct AddRolloutRequest {
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub payload_checksum: Option<String>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub artifact_type: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
     pub metadata: Option<Value>,
 }
 
@@ -741,6 +743,10 @@ pub struct RolloutRecordDto {
     pub payload_size: Option<i64>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub payload_checksum: Option<String>,
+    /// User-defined semantic artifact category (e.g. `"excel_grade_screenshot"`),
+    /// orthogonal to `content_type`. A first-class, filterable column.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub artifact_type: Option<String>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub metadata: Option<Value>,
 }

crates/lance-context-core/src/api_impl.rs

@@ -458,6 +458,7 @@ fn rollout_record_from_add_request(r: &AddRolloutRequest) -> RolloutRecord {
         binary_payload: r.binary_payload.clone(),
         payload_size: r.payload_size,
         payload_checksum: r.payload_checksum.clone(),
+        artifact_type: r.artifact_type.clone(),
         metadata: r.metadata.clone(),
     }
 }
@@ -497,6 +498,7 @@ fn rollout_record_to_dto(r: RolloutRecord) -> RolloutRecordDto {
         binary_payload: r.binary_payload,
         payload_size: r.payload_size,
         payload_checksum: r.payload_checksum,
+        artifact_type: r.artifact_type,
         metadata: r.metadata,
     }
 }

crates/lance-context-core/src/rollout.rs

@@ -84,8 +84,16 @@ pub struct RolloutRecord {
     pub binary_payload: Option<Vec<u8>>,
     pub payload_size: Option<i64>,
     pub payload_checksum: Option<String>,
+    /// User-defined semantic category of an artifact, e.g.
+    /// `"excel_grade_screenshot"`. Orthogonal to `content_type`, which is the
+    /// transport/media type (e.g. `"image/png"`): `content_type` says how to
+    /// decode the bytes, `artifact_type` says what the artifact *means*. A
+    /// first-class column so it can be filtered / grouped-by / projected
+    /// without materializing the free-form `metadata` JSON.
+    pub artifact_type: Option<String>,
     /// Harness metadata — the open-ended escape hatch, for genuinely
-    /// unstructured fields only (e.g. an artifact's `filename` / `artifact_type`).
+    /// unstructured fields only (e.g. an artifact's `filename`). Semantic
+    /// categories that you filter/group-by belong in `artifact_type` instead.
     pub metadata: Option<Value>,
 }
 

crates/lance-context-core/src/rollout_store.rs

@@ -423,6 +423,7 @@ impl RolloutStore {
         let mut binary_payload_builder = LargeBinaryBuilder::new();
         let mut payload_size_builder = Int64Builder::new();
         let mut payload_checksum_builder = StringBuilder::new();
+        let mut artifact_type_builder = StringBuilder::new();
         let mut metadata_builder = LargeStringBuilder::new();
 
         for record in records {
@@ -482,6 +483,7 @@ impl RolloutStore {
             }
             payload_size_builder.append_option(record.payload_size);
             payload_checksum_builder.append_option(record.payload_checksum.as_deref());
+            artifact_type_builder.append_option(record.artifact_type.as_deref());
             match &record.metadata {
                 Some(metadata) => metadata_builder.append_value(metadata.to_string()),
                 None => metadata_builder.append_null(),
@@ -589,6 +591,10 @@ impl RolloutStore {
             "payload_checksum".to_string(),
             Arc::new(payload_checksum_builder.finish()),
         );
+        arrays_by_name.insert(
+            "artifact_type".to_string(),
+            Arc::new(artifact_type_builder.finish()),
+        );
         if include_metadata {
             arrays_by_name.insert("metadata".to_string(), Arc::new(metadata_builder.finish()));
         }
@@ -682,6 +688,7 @@ pub fn rollout_schema() -> Schema {
         binary_field,
         Field::new("payload_size", DataType::Int64, true),
         Field::new("payload_checksum", DataType::Utf8, true),
+        Field::new("artifact_type", DataType::Utf8, true),
         Field::new("metadata", DataType::LargeUtf8, true),
     ];
 
@@ -767,6 +774,7 @@ fn batch_to_rollout_records(batch: &RecordBatch) -> LanceResult<Vec<RolloutRecor
     let binary_payload_array = column_as_optional::<LargeBinaryArray>(batch, "binary_payload");
     let payload_size_array = column_as_optional::<Int64Array>(batch, "payload_size");
     let payload_checksum_array = column_as_optional::<StringArray>(batch, "payload_checksum");
+    let artifact_type_array = column_as_optional::<StringArray>(batch, "artifact_type");
     let metadata_array = column_as_optional::<LargeStringArray>(batch, "metadata");
 
     let mut results = Vec::with_capacity(batch.num_rows());
@@ -855,6 +863,7 @@ fn batch_to_rollout_records(batch: &RecordBatch) -> LanceResult<Vec<RolloutRecor
                 }
             }),
             payload_checksum: optional_string(payload_checksum_array, row),
+            artifact_type: optional_string(artifact_type_array, row),
             metadata,
         });
     }
@@ -997,6 +1006,7 @@ mod tests {
             binary_payload: None,
             payload_size: None,
             payload_checksum: None,
+            artifact_type: None,
             metadata: Some(json!({"harness": "verifiers"})),
         }
     }
@@ -1032,7 +1042,8 @@ mod tests {
             binary_payload: Some(bytes.to_vec()),
             payload_size: Some(bytes.len() as i64),
             payload_checksum: Some("sha256:cafef00d".to_string()),
-            metadata: Some(json!({"filename": "trace.bin", "artifact_type": "trace"})),
+            artifact_type: Some("excel_grade_screenshot".to_string()),
+            metadata: Some(json!({"filename": "trace.bin"})),
         }
     }
 
@@ -1074,6 +1085,7 @@ mod tests {
         // round-trip.
         assert_eq!(actual.payload_size, expected.payload_size);
         assert_eq!(actual.payload_checksum, expected.payload_checksum);
+        assert_eq!(actual.artifact_type, expected.artifact_type);
         assert_eq!(actual.metadata, expected.metadata);
     }
 

crates/lance-context-server/src/routes/rollouts.rs

@@ -443,6 +443,7 @@ fn rollout_record_from_add_request(r: &AddRolloutRequest) -> RolloutRecord {
         binary_payload: r.binary_payload.clone(),
         payload_size: r.payload_size,
         payload_checksum: r.payload_checksum.clone(),
+        artifact_type: r.artifact_type.clone(),
         metadata: r.metadata.clone(),
     }
 }
@@ -482,6 +483,7 @@ fn rollout_record_to_dto(r: RolloutRecord) -> RolloutRecordDto {
         binary_payload: r.binary_payload,
         payload_size: r.payload_size,
         payload_checksum: r.payload_checksum,
+        artifact_type: r.artifact_type,
         metadata: r.metadata,
     }
 }

crates/lance-context/src/lib.rs

@@ -12,9 +12,9 @@ pub use lance_context_core::{
 pub use lance_context_api::{
     AddRecordRequest, AddRecordsResponse, AddRolloutRequest, AddRolloutsResponse, CompactRequest,
     CompactResponse, CompactStatsResponse, ContextError, ContextResult, ContextStoreApi,
-    DeleteRecordResponse, RecordDto, RelationshipDto, RetrieveRequest, RetrieveResponse,
-    RetrieveResultDto, RolloutRecordDto, RolloutStoreApi, SearchResultDto, UpsertRecordRequest,
-    UpsertRecordResponse,
+    CreateRolloutStoreRequest, DeleteRecordResponse, RecordDto, RelationshipDto, RetrieveRequest,
+    RetrieveResponse, RetrieveResultDto, RolloutRecordDto, RolloutStoreApi, SearchResultDto,
+    UpsertRecordRequest, UpsertRecordResponse,
 };
 
 #[cfg(feature = "remote")]

python/Cargo.toml

@@ -11,6 +11,7 @@ crate-type = ["cdylib"]
 
 [dependencies]
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
+lance-context = { path = "../crates/lance-context", features = ["remote"] }
 lance-context-api = { path = "../crates/lance-context-api" }
 lance-context-client = { path = "../crates/lance-context-client" }
 lance-context-core = { path = "../crates/lance-context-core" }

python/python/lance_context/__init__.py

@@ -6,6 +6,8 @@
     ContextNamespace,
     EmbeddingProvider,
     RemoteContext,
+    RemoteRolloutStore,
+    RolloutStore,
     __version__,
 )
 from .embeddings import (  # pyright: ignore[reportMissingImports]
@@ -19,5 +21,7 @@
     "EmbeddingProvider",
     "MultiModalEmbeddingProvider",
     "RemoteContext",
+    "RemoteRolloutStore",
+    "RolloutStore",
     "__version__",
 ]