|
| 1 | +from __future__ import annotations |
| 2 | + |
| 3 | +import asyncio |
| 4 | +from typing import List, Optional, Tuple |
| 5 | + |
| 6 | +from moss_core import ( |
| 7 | + AddDocumentsOptions as _RustAddDocumentsOptions, |
| 8 | + DocumentInfo, |
| 9 | + GetDocumentsOptions, |
| 10 | + MutationOptions, |
| 11 | + QueryOptions, |
| 12 | + SearchResult, |
| 13 | + SessionIndex as _RustSessionIndex, |
| 14 | + PushIndexResult, |
| 15 | +) |
| 16 | + |
| 17 | + |
| 18 | +class SessionIndex: |
| 19 | + """ |
| 20 | + A local in-session index for real-time indexing and querying. |
| 21 | +
|
| 22 | + All operations (add_docs, delete_docs, query) run entirely in-memory |
| 23 | + with no cloud round trips. Call push_index() at session end to persist |
| 24 | + the index to the cloud for bookkeeping and future retrieval. |
| 25 | +
|
| 26 | + Usage telemetry is tracked and reported automatically in the background. |
| 27 | +
|
| 28 | + Example: |
| 29 | + ```python |
| 30 | + # Auto-loads from cloud if the index exists, starts fresh if not |
| 31 | + session = await client.session(index_name="session-abc") |
| 32 | +
|
| 33 | + await session.add_docs([DocumentInfo(id="1", text="Customer asked about billing")]) |
| 34 | + results = await session.query("billing question") |
| 35 | +
|
| 36 | + result = await session.push_index() |
| 37 | + # optionally: await client.get_job_status(result.job_id) |
| 38 | + ``` |
| 39 | + """ |
| 40 | + |
| 41 | + def __init__(self, name: str, model_id: str, _inner: "_RustSessionIndex") -> None: |
| 42 | + self._model_id = model_id |
| 43 | + self._inner = _inner |
| 44 | + |
| 45 | + @classmethod |
| 46 | + def _create( |
| 47 | + cls, |
| 48 | + name: str, |
| 49 | + model_id: str, |
| 50 | + project_id: str, |
| 51 | + project_key: str, |
| 52 | + client_id: Optional[str] = None, |
| 53 | + ) -> "SessionIndex": |
| 54 | + inner = _RustSessionIndex(name, model_id, project_id, project_key, client_id) |
| 55 | + return cls(name, model_id, inner) |
| 56 | + |
| 57 | + @property |
| 58 | + def name(self) -> str: |
| 59 | + """The index name.""" |
| 60 | + return self._inner.name |
| 61 | + |
| 62 | + @property |
| 63 | + def doc_count(self) -> int: |
| 64 | + """Number of documents in the local session index.""" |
| 65 | + return self._inner.doc_count |
| 66 | + |
| 67 | + async def add_docs( |
| 68 | + self, |
| 69 | + docs: List[DocumentInfo], |
| 70 | + options: Optional[MutationOptions] = None, |
| 71 | + ) -> Tuple[int, int]: |
| 72 | + """ |
| 73 | + Add or update documents in the local session index. |
| 74 | +
|
| 75 | + Embeddings are generated locally via Rust core — no cloud round trip. |
| 76 | +
|
| 77 | + Args: |
| 78 | + docs: Documents to add. When using model_id='custom', each doc |
| 79 | + must have .embedding set. |
| 80 | + options: Mutation options (e.g. upsert behavior). |
| 81 | +
|
| 82 | + Returns: |
| 83 | + Tuple of (added_count, updated_count). |
| 84 | + """ |
| 85 | + rust_opts = None |
| 86 | + if options is not None: |
| 87 | + upsert = bool(options.upsert) if options.upsert is not None else True |
| 88 | + rust_opts = _RustAddDocumentsOptions(upsert=upsert) |
| 89 | + if self._model_id == "custom": |
| 90 | + embeddings = self._get_custom_embeddings(docs) |
| 91 | + return await asyncio.to_thread( |
| 92 | + self._inner.add_docs, |
| 93 | + docs, |
| 94 | + embeddings, |
| 95 | + rust_opts, |
| 96 | + ) |
| 97 | + return await asyncio.to_thread(self._inner.add_docs_text, docs, rust_opts) |
| 98 | + |
| 99 | + async def delete_docs(self, doc_ids: List[str]) -> int: |
| 100 | + """ |
| 101 | + Delete documents from the local session index by their IDs. |
| 102 | +
|
| 103 | + Returns: |
| 104 | + Number of documents deleted. |
| 105 | + """ |
| 106 | + return await asyncio.to_thread(self._inner.delete_docs, doc_ids) |
| 107 | + |
| 108 | + async def get_docs( |
| 109 | + self, |
| 110 | + options: Optional[GetDocumentsOptions] = None, |
| 111 | + ) -> List[DocumentInfo]: |
| 112 | + """Retrieve documents from the local session index.""" |
| 113 | + return await asyncio.to_thread(self._inner.get_docs, options) |
| 114 | + |
| 115 | + async def query( |
| 116 | + self, |
| 117 | + query: str, |
| 118 | + options: Optional[QueryOptions] = None, |
| 119 | + ) -> SearchResult: |
| 120 | + """ |
| 121 | + Perform a semantic search over the local session index. |
| 122 | +
|
| 123 | + Runs entirely in-memory (~1-10ms). No cloud call. |
| 124 | +
|
| 125 | + Args: |
| 126 | + query: The search query text. |
| 127 | + options: Query options (top_k, alpha, embedding, filter). Example filter: |
| 128 | + QueryOptions(filter={"$and": [ |
| 129 | + {"field": "type", "condition": {"$eq": "faq"}}, |
| 130 | + {"field": "priority", "condition": {"$gt": "5"}}, |
| 131 | + ]}) |
| 132 | +
|
| 133 | + Returns: |
| 134 | + SearchResult with scored documents. |
| 135 | + """ |
| 136 | + top_k = getattr(options, "top_k", None) |
| 137 | + top_k = top_k if top_k is not None else 5 |
| 138 | + alpha = getattr(options, "alpha", None) |
| 139 | + alpha = alpha if alpha is not None else 0.8 |
| 140 | + query_embedding = getattr(options, "embedding", None) |
| 141 | + filter = getattr(options, "filter", None) |
| 142 | + |
| 143 | + if query_embedding is None: |
| 144 | + if self._model_id == "custom": |
| 145 | + raise ValueError( |
| 146 | + "This session uses custom embeddings. " |
| 147 | + "Provide a query embedding via QueryOptions.embedding." |
| 148 | + ) |
| 149 | + return await asyncio.to_thread( |
| 150 | + self._inner.query_text, |
| 151 | + query, |
| 152 | + top_k, |
| 153 | + alpha, |
| 154 | + filter, |
| 155 | + ) |
| 156 | + |
| 157 | + return await asyncio.to_thread( |
| 158 | + self._inner.query, |
| 159 | + query, |
| 160 | + top_k, |
| 161 | + list(query_embedding), |
| 162 | + alpha, |
| 163 | + filter, |
| 164 | + ) |
| 165 | + |
| 166 | + async def push_index(self) -> PushIndexResult: |
| 167 | + """ |
| 168 | + Push the local session index to the cloud. |
| 169 | +
|
| 170 | + Sends all documents with their locally-computed embeddings to the |
| 171 | + backend. The cloud index is created or replaced if one already exists |
| 172 | + with the same name. No server-side re-embedding occurs. |
| 173 | +
|
| 174 | + Returns: |
| 175 | + PushIndexResult with job_id and status. |
| 176 | + """ |
| 177 | + return await asyncio.to_thread(self._inner.push_index) |
| 178 | + |
| 179 | + async def _get_embedding_service(self) -> None: |
| 180 | + if self._model_id != "custom": |
| 181 | + await asyncio.to_thread(self._inner.load_model) |
| 182 | + |
| 183 | + def _get_custom_embeddings(self, docs: List[DocumentInfo]) -> List[List[float]]: |
| 184 | + missing = [doc.id for doc in docs if not getattr(doc, "embedding", None)] |
| 185 | + if missing: |
| 186 | + raise ValueError( |
| 187 | + f"Documents missing .embedding for custom model: {missing}. " |
| 188 | + "All documents must have .embedding set when using model_id='custom'." |
| 189 | + ) |
| 190 | + return [list(doc.embedding) for doc in docs] # type: ignore[arg-type] |
0 commit comments