Crate llm_kernel

Expand description

§llm-kernel

Foundation library for Rust AI-native applications.

Provides a composable, feature-gated set of modules for building LLM-powered tools, agents, and servers:

Feature	Module	Description
`provider`	`provider`	Provider catalog, model descriptors, pricing — default
`client-async`	[`llm`]	Async LLM client (OpenAI, Anthropic) with SSE streaming
`discovery`	[`discovery`]	Dynamic model discovery (models.dev, Ollama, OpenAI-compat)
`secrets`	[`secrets`]	SecretVault — dotenv-style credential management
`store`	[`store`]	SQLite init helpers (WAL, PRAGMA, schema versioning)
`config`	[`config`]	TOML config loader with auto-create from template
`graph`	[`graph`]	Knowledge graph — SQLite, FTS5, smart recall, BFS traversal
`mcp`	[`mcp`]	MCP server framework — JSON-RPC 2.0, stdio transport
`tokens`	[`tokens`]	Token estimation with Unicode-script heuristics
`install`	[`install`]	AI tool installation wizard (Claude, Cursor, Copilot, etc.)
`search`	[`search`]	Hybrid search with Reciprocal Rank Fusion
`embedding`	[`embedding`]	Embedding provider trait + cosine similarity
`telemetry`	[`telemetry`]	Telemetry framework — enum-gated events, no PII
`safety`	[`safety`]	Secret masking, error classification, output sanitization

The prelude module re-exports the most commonly used types:

use llm_kernel::prelude::*;