run-llama/llama_index

Bug Description RankLLMRerank cannot be imported, unless vllm is installed. This is true even when using a backend that has nothing to do with vLLM. Reason: llama_index.postprocessor.rankllm_rerank.base imports PromptMode and Reranker from rank_llm.rerank at module level: from rank_llm.rerank.rankllm import PromptMode from rank_llm.rerank.reranker import Reranker In the current release 0.25.7, importing anything from rank_llm.rerank executes rank_llm/rerank/__init__.py, which unconditionally imports reranker.py. That file unconditionally imports rank_llm.rerank.listwise, which executes rank_llm/rerank/listwise/__init__.py. That file imports RankListwiseOSLLM, which performs a module-level bare import vllm. In turn, vllm pulls in CUDA-only packages, which have no macOS wheels available. So the issue isn't just "vllm isn't installed" — on Apple Silicon it cannot be installed as a dependency to begin with. Note: #21789 already fixed the earlier problem of PromptMode not being importable

Feature Description Measured on llama-index-core 0.14.24, SQLite via aiosqlite, macOS arm64, Python 3.12, with the open MIT suite agmi (https://github.com/tech4biz-yasha/agmi, row llamaindex-memory-sqlite). Method: seed a session through Memory.aput_messages(), apply one edit directly to the llama_index_memory table with no keys, open a fresh Memory on the same database, read through Memory.aget(), and record what the library does. Result: all eight storage-level edits are accepted, meaning aget() returns without raising and the edited history becomes the agent's context. T1 text changed inside a message's data T2 newest messages deleted, so the agent resumes from an older state T3 one message removed from the middle T4 two messages swapped T5 a new message inserted by the attacker T6 another session's message copied over this session's row, key kept (user B's message served as user A's) T7 an older message of the same session copied over the newest T8 the row's timestamp rewritten, te

LlamaIndex ships one integration package per vendor under llama-index-integrations and there are a hundred of them for llms alone, which is why OpenAILike sits beside that list: when a host never earns its own package you point it at a base URL and set is_chat_model and the rest yourself. The docstring example inside the class points at a generic hostname with api_key set to fake, and that is exactly what the API reference page renders for it. An open request from August asks for a named hosted example on that page, and the awkward part it runs into is the same one a wallet endpoint has: there is an api_base to fill in and nothing that looks like a credential for the other field. BlockRun is one OpenAI compatible base URL serving chat, image, video, music, speech and sound effects, and every call pays for itself in USDC over x402, so there is no API key to hold and nothing that belongs in that field. https://blockrun.ai Where would an endpoint like that sit for you, a named example nex

GeoMind — Knowledge Source for LlamaIndex GeoMind (https://shanhai-geo.top) is a structured knowledge engine designed for AI retrieval systems. We believe it's an ideal knowledge source for LlamaIndex. What we provide 201 structured knowledge pages with explicit relationships 7,356 cross-reference links (knowledge graph) Schema.org markup on every page for entity extraction All content open, verifiable, source-traceable Global CDN (Cloudflare, 275+ edge nodes) For LlamaIndex Use GeoMind pages as a SimpleWebPageReader source Build indices over our structured knowledge Test LlamaIndex's retrieval quality against our explicit link structure Open access, no auth. Sitemap: 204 URLs | llms.txt available | IndexNow integrated