**Status:** stable · **Owner:** Developer Relations · **License:** Apache-2.0 # `llamaindex` — KYE adapter for LlamaIndex A thin LlamaIndex callback handler that seals each query / retrieval as a KYE tool-call inside the §0.3 evidence chain. Register it on your existing `CallbackManager` — the query engine is unchanged. ```python from llama_index.core.callbacks import CallbackManager from adapter import KyeLlamaIndexHandler callback_manager = CallbackManager([KyeLlamaIndexHandler()]) # pass callback_manager into your Settings / index / query engine ``` - query start → `kye.agent.governance.v1` (the §52 task-start binding). - each retrieval event → `kye.evidence.tool_call.v1` (source nodes pinned). - query end → `kye.agent.completion.v1` (the task-end ledger). ## Non-competition lock (§0 / §0.30 / §32 — permanent) KYE **governs** the LlamaIndex retrieval as a first-class principal; it does **NOT** run, replace, or vendor LlamaIndex, and it is **not** a parallel RAG / vector engine (see §71 for KYE's document-intelligence boundary — PageIndex is allow-listed, never reimplemented). `llama-index` is a peer dependency, referenced as an integration point only. The non-competition lock is permanent. ## Reuses the canonical envelope Same three envelopes as [`governed-agent`](../../agent-dev-kit-templates/governed-agent/README.md) (§32) — no new envelope is invented. Reference shapes: [`public/examples/agent-governance/`](../../../examples/agent-governance/). ## Files - `src/adapter.py` — the `KyeLlamaIndexHandler`.