llm-integration

A multi-provider language-model integration for applications that can use Gemini, OpenAI, Anthropic, or a local Ollama model. It supports optional web grounding and streaming responses.

In plain words
What is it for?
Use it to generate or stream text through a selected provider, optionally ground answers with web search, and support users who bring their own model keys.
Why use it?
It gives the application one interface for different model providers while handling provider selection, retries, caching, and user-supplied API keys.

Cursor rule for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/eliornl/rolemule/llm-integration
Clone the repo
git clone --depth 1 https://github.com/eliornl/rolemule

Made for: Claude Code.

Per session 0 Nothing until a file matches its globs; then the whole rule loads.
When invoked 1,942 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.01942
Opus 5 $0.00000 $0.00971
Sonnet 5 $0.00000 $0.00388
Haiku 4.5 $0.00000 $0.00194

Measured 2d ago against content hash e70d0026c5d0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

llm-integration scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/rules/llm-integration.mdc · 155 lines

How it starts

The opening of the file, as written. The whole thing — 155 lines — stays where its author put it; the contents beside it link to each section on GitHub.

LLM Integration Patterns

Architecture

utils/llm/
  client.py          # LLMClient facade (cache + retry + route by provider=); generate + generate_stream
  availability.py    # resolve_user_llm_context / CFG_6001 readiness
  registry.py        # provider name → factory
  speak_stream.py    # SpeakFieldStreamer — progressive JSON "speak" extraction
  providers/
    gemini.py        # Google AI Studio + optional Vertex + Google Search grounding (+ stream)
    openai.py        # Chat Completions + Responses web_search when grounded (+ stream)
    anthropic.py     # Messages API + web_search server tool when grounded (+ stream)
    ollama.py        # Local /api/chat (no tools / no key) (+ stream)
utils/llm_client.py  # backward-compatible re-exports
utils/llm_context.py # require_user_llm_context() for API gates

Prefer get_llm_client() for new code. get_gemini_client() remains an alias.

from utils.llm_client import get_llm_client

llm = await get_llm_client()
response = await llm.generate(
    prompt=prompt,
    system=system_message,
    temperature=0.7,
    max_tokens=16000,
    user_api_key=user_api_key,   # decrypted BYOK for the resolved provider
    provider=llm_provider,       # required for multi-user correctness
    model=user_model,            # None → provider default
)
text = response.get("response", "")

Stable response shape: { "response": str, "done": bool, "model"?: str, "filtered"?: bool, "from_cache"?: bool }.

Streaming (generate_stream)

Use for progressive UI (Practice Interview interviewer speak). Not cached and not retried like generate().

async for chunk in llm.generate_stream(
    prompt=prompt,
    system=system_message,
    temperature=0.55,
    max_tokens=16000,
    user_api_key=user_api_key,
    provider=llm_provider,
    model=user_model,
):
    # chunk is a text delta (str)
    ...

Providers implement generate_stream on Gemini / OpenAI / Anthropic / Ollama. On stream failure, callers should fall back to non-stream generate().

Read the full file on GitHub · 155 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 155 lines · 0 tokens per session scan A e70d0026c5d0

Subscribe to this mod's changes

llm-integration is a cursor rule published in the GitHub repository eliornl/rolemule (37 stars, last pushed 3d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,942 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.