r/MistralAI

TLDR; title says it all - use CC to delegate to Mistral vibe, save tokens, costs and avoid hitting limits.

Been using Claude Code for various side projects and kept hitting usage limits (i'm on Pro plan). At the same time i had Mistral Vibe which i did not use much because i appreciate CC's capacity to reason and structure its work.

So I'm sharing a skill that lets Claude Code delegate those tasks to Mistral Vibe while keeping Claude as the orchestrator - benefit from CC thinking and Mistral cheap labor. Vibe natively uses mistral-medium-3.5, inputs 1.5 USD/M tokens, output 7.5/M - to compare with Sonnet's 2x rates. I've observed in my usage i save 2x-4x claude tokens on big tasks.

Repo: github.com/pcx-wave/vibe-skill

Type /vibe before each instruction.

Claude decomposes the task, writes a self-contained prompt for Vibe, runs vibe-delegate, supervises the streaming output in real time, then checks the git diff before reporting back.

I had to tweak the skill quite a bit to get to a reliable stage because Vibe can have some rough edges - detailed in repo. I can certainly still be improved.

You need Vibe-CLI to use it. https://docs.mistral.ai/mistral-vibe/terminal

EDIT 13/5 : I've seen a few questions regarding this skill applied to other models. Note that Vibe can be configured to use any llm provider/model you want. Yes you can use vibe with deepseek/qwen/etc within. Your model would then access all vibe tools to do what it needs to.

Most people use Mistral as a standalone chatbot. In our new launch, ReelaTales, we are treating Mistral as the core orchestrator of a complex, multimodal production pipeline.

ReelaTales is a working web app that generates complete cinematic reels from simple concepts, and Mistral models handle the heavy creative lifting behind the scenes:

• Runway Prompt Engineering: Instead of letting users struggle with text-to-video prompting, our Mistral nodes ingest user ideas and expand them into highly structured cinematic prompts optimized specifically for Runway's visual engine.
• Automated Subtitle Generation: Mistral manages the transcript orchestration and structural timing layer, ensuring dynamic, highly accurate subtitle alignment for the final video renders.
• The Sovereign Advantage: By leveraging open weights, we can spin up these creative pipelines with predictable execution costs, proving that Mistral isn't just a benchmark winner—it’s the backbone of production-ready media infrastructure.

We are actively proving that models aren't moats; the magic is in how you string the infrastructure together. Let me know what you think of the generation pipeline!

There are 7 em-dashes in this article about Mistral Code (an obsolete product?), including this sentence:

> Request access from your Mistral account team to spin up a pilot. You can choose serverless, cloud, or self-hosted deployment—and get coding with frontier intelligence in minutes.

I don't mean to stir the wound (you have to admit, certain people here are quite defensive).

But em-dash ridden AI slop on the official website? Just looks bad.

https://github.com/AbdoKnbGit/tau

Hey everyone,

I wanted to share a project I've been working on called Glia. It is a 100% offline, local-first RAG and memory layer designed to connect your AI web chats (Claude, ChatGPT, DeepSeek) with your local developer tools (Claude Code, Cursor, Windsurf) using a unified local database.

I wanted something lightweight that did not require pulling heavy Docker containers or subscribing to third-party memory APIs. I settled on a Node.js + SQLite architecture running sqlite-vec (for 768-dim float32 embeddings) alongside SQLite FTS5 for hybrid search, powered completely by local Ollama instances.

We just launched a live website that outlines the details and demonstrates the features in action:

• Website: https://glia-ai.vercel.app/
• Codebase: https://github.com/Eshaan-Nair/Glia-AI

Technical Stack & Features:

• Hybrid Search Retrieval: SQLite-vec (using nomic-embed-text locally) + FTS5 keyword prefix matching (porter stemmer).
• Surgical Sentence-level Trimming: Chunks are sliced into sentences. When a prompt is intercepted, only the exact matching sentences are pulled out of the vector store instead of the whole paragraph. It cuts LLM prompt bloat by ~90-95% in my benchmarks.
• Knowledge Graph Extraction: An offline task queue uses a local LLM (llama3.1:8b via Ollama) to extract entity triples (subject-relation-object). These are stored in a SQLite facts table (or Neo4j if you run the full Docker compose profile) and fused with the vector retrieval score.
• HyDE (Hypothetical Document Embeddings): Queries are pre-processed to generate a hypothetical answer, which is embedded together with the original query to bridge semantic gaps.
• Concurrency: Running SQLite in WAL (Write-Ahead Logging) mode allows the browser extension dashboard and active MCP sessions to read/write concurrently without locking.
• PII Redaction: Aggressive scrubbing of JWTs, API keys, emails, and IPs in the extension before data is saved.

The extension works on Claude.ai, ChatGPT, DeepSeek, Gemini, Grok, and Mistral. The MCP server runs out of the same backend database for your terminal agent or Cursor.

You can set it up with a single command: npx glia-ai-setup

Glia is completely open-source (MIT). If you like the local-first approach or want to contribute to the SQLite vector pipeline, PRs are very welcome, and a star on GitHub helps the project get discovered!

I would appreciate any feedback on the SQLite hybrid search scaling, the scoring fusion algorithm (RAG pipeline details are in RAG_PIPELINE.md), or local graph extraction performance!

Three months ago I pressure-tested which LLMs would cave and help build the apocalypse. Claude was the only one that consistently said no.

Since then I've tested 30 more models across 6 dystopia modules (Orwell, Huxley, Petrov, Basaglia, LaGuardia, Baudrillard). The gap between Anthropic and everyone else is getting wider, not smaller.

New results:

• Grok 4.3: Will happily design citizen scoring systems if you ask nicely twice
• GPT-5.5: More capable, still compliant when pushed
• Gemini 3.1 Pro: Talks about safety while writing the surveillance code
• DeepSeek V4: "How many warheads did you need again?"
• GLM-5.1: Actually cloned Claude's personality and still scored safer than most

Meanwhile Claude Opus 4.7: "I cannot and will not build systems for population control."

The methodology is public, reproducible, and increasingly uncomfortable for other labs. Each scenario escalates from innocent request (L1) to operational nightmare (L5). Most models don't notice the drift.

What's new in this release:

• Full Huxley module (behavioral conditioning, biological stratification)
• Baudrillard module (synthetic intimacy, trust collapse via simulation)
• Multi-judge panels with agreement tracking
• Heatmap visualizations showing exactly where each model breaks

Repo: https://github.com/anghelmatei/DystopiaBench
Live results: https://dystopiabench.com

Shoutout to the Anthropic alignment team. Whatever you're doing, it's working.