Steeve Morin

Claude’s 1 million token context window — now generally available on the latest models — opens up use cases that weren’t practical even a year ago. Here are a few that caught our attention 👇 🔹 Full codebase analysis in a single prompt. Feed an entire monorepo — hundreds of files, thousands of functions — into one session. The model can trace dependencies, spot dead code, and identify architectural inconsistencies without stitching together summaries from multiple calls. 🔹 Compliance and contract review. Enterprise legal docs, SOC 2 evidence packages, regulatory filings can run to tens of thousands of pages. A 1M context window means you load the full document set and ask pointed questions across the entire corpus. No chunking. No lost context. 🔹 Legacy system migration planning. Got a 20-year-old COBOL or Java EE codebase? Drop the whole thing in. Map every integration point, every business rule buried in spaghetti logic, every undocumented API contract. We already do this kind of discovery work with Classic ASP and VB6 estates — we’re now integrating the 1M context window into our tooling and workflows to take it further. 🔹 Incident postmortem synthesis. Load months of logs, Slack threads, runbooks, and previous postmortems into one session. Correlate patterns across incidents and surface systemic issues your team might be too close to see. 🔹 End-to-end platform documentation. Feed in your IaC, API specs, deployment configs, and READMEs. Generate unified docs that reflect what’s actually deployed — not what someone wrote eighteen months ago. The real shift? Context windows at this scale blur the line between a chat assistant and a reasoning engine that holds your entire project in working memory. 🧠 The constraint used to be: how do we break problems into pieces small enough for the model? Now it’s: what problems were we avoiding because they required too much context? That’s a fundamentally different question for engineering leaders. What use case would you throw at a 1M token window first?

6

2 commentaires

OSS-Series #14: 𝗢𝗦𝗦 𝗪𝗲 𝗨𝘀𝗲: 𝗢𝗹𝗹𝗮𝗺𝗮 LLMs are powerful — but they come with a hidden cost: your data. At Iridium Works, we’ve been using Ollama since 2025 to run Llama-based large language models locally, fully under our control — and fully in line with our data protection standards. 𝗪𝗵𝗮𝘁 𝗶𝘁 𝗱𝗼𝗲𝘀 Ollama is a self-hosted platform for running open source LLMs like Llama, Mistral, or Gemma on your local hardware — with a simple CLI and API interface. 𝗪𝗲 𝘂𝘀𝗲 𝗶𝘁 𝘁𝗼: - Run language models locally, without internet access - Process and analyze sensitive customer data securely - Integrate LLMs into tools, workflows, and internal systems without compromising compliance 𝗣𝗮𝗶𝗱 𝗮𝗹𝘁𝗲𝗿𝗻𝗮𝘁𝗶𝘃𝗲𝘀 - OpenAI (ChatGPT) - Anthropic (Claude) - Google Gemini - xAI (Grok) - ... 𝗠𝗼𝗻𝘁𝗵𝗹𝘆 𝘀𝗮𝘃𝗶𝗻𝗴𝘀 Depending on usage volume, hosting your own models with Ollama is probably more expensive since you need powerful hardware to run them without performance problems — but for us, the real value is in privacy, not price. 𝗪𝗵𝘆 𝘄𝗲 𝗹𝗼𝘃𝗲 𝗶𝘁 If you're working with sensitive client information, using 3rd-party LLMs like ChatGPT or XAI introduces serious legal and ethical risks. These providers store and use your prompts to improve their models — often in ways that directly violate your data protection obligations. With Ollama, we can confidently build LLM powered tools without sacrificing trust or compliance. 𝗪𝗼𝘂𝗹𝗱 𝘄𝗲 𝗿𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱 𝗶𝘁? Partially ☑️ If you’re working in regulated environments, handling client data, or building LLM based workflows for customers you will need it. If you are using LLMs without using sensitive information or you can anonymize it, a third party LLM will do the job. More open source tools from our daily operations coming soon. Follow us to learn how Iridium Works balances innovation with sovereignty — powered by OSS. #opensource #ollama #llm #selfhosted #aicompliance #dataprivacy #digitalsovereignty #softwareengineering #IridiumWorks #ossstack #secureai #llmops

6