Open Source Innovation Collaboration

OpenAI, Anthropic, and Block quietly did something important yesterday: they donated core pieces of their agent technology into a new neutral nonprofit called the Agentic AI Foundation (AAIF), established under the Linux Foundation. The contributions include AGENTS.md, Anthropic’s Model Context Protocol (MCP), and Block’s open source “goose” framework. Together, these tools create the early architecture for interoperable agents. AGENTS.md is the simplest part and also the most disruptive. It is a plain Markdown file that sits in the root of a code repository and acts like a contract for any agent that touches that code. It tells the agent how to build, test, lint, secure, and operate the project. Developers do not have to rely on tool-specific documentation or proprietary wrappers. The rules live in one place. Any supported agent knows what to do. This matters because today, every system has its own toolchain, syntax, connection method, and expectations. The same project behaves differently when a developer switches from one AI assistant to another. Workflows break. Output becomes inconsistent. Risk increases. AGENTS.md standardizes this. It creates predictable behavior. It cuts integration complexity. It turns codebases into agent-ready environments that do not depend on a single vendor. The foundation that now holds these tools matters as much as the tools themselves. The Linux Foundation has a long track record of managing open standards that reshape industries. OpenAI and Anthropic have a strong record of influencing the direction of agent development. The industry finally has the beginnings of a shared vocabulary for how agents should behave. In practice, if agents follow agreed protocols, the market becomes less about individual models and more about the quality of the workflows built on top of them. The competition shifts to orchestration, trust, and domain expertise. It is still early. These standards will only work if the ecosystem adopts them and if enterprises build governance frameworks that align with them. Yesterday’s announcement did not dominate the news cycle. It should have. If everyone in your organization is using the words “agent” and “agentic,” this is worth your attention.

The most powerful asset at CrewAI isn't our technology per se — it's our open-source community. Why? Because an engaged community creates flywheels that are incredibly hard to replicate: • Innovation Flywheel: Contributors constantly refine the product, accelerating our pace. • Trust Flywheel: Transparency in open-source builds unmatched credibility and loyalty. • Growth Flywheel: Word-of-mouth from passionate users expands our audience organically. Once you've built a strong open-source community, endless opportunities emerge to serve and support it. For example, we recently launched jobs.crewai.com — a dedicated job board connecting companies and professionals passionate about multi-agent AI. That now has HUNDREDS of signed-up CrewAI Engineers from different countries that we are pulling into jobs and opportunities for customers, some already doing work. Building open-source isn't just about giving back—it's about cultivating a thriving ecosystem where everyone grows together. Community-driven innovation is unstoppable. Have you experienced the power of open-source in your organization?

Here’s everything you need to know about Open Source Toolkit for building AI Agents. I have been exploring what it really takes to build practical AI agents, and the open-source ecosystem has come a long way. There are now powerful tools for every layer of the stack: • Browser automation to navigate and interact with the web • Document processing with tools like DocOwl2 • Research frameworks such as GPT Researcher and Local Deep Research • Vertical agents built for specific workflows • Computer control with Open Interpreter and Self Operating Computer • Voice interfaces powered by Parakeet v2 and ChatTTS • Memory, evaluation and monitoring to refine performance The exciting part is how these tools can be combined to create agents that are not just demos but actually usable in real-world workflows If you are building in this space, this open-source toolkit is worth exploring. It might save you weeks of work and spark ideas you did not think were possible #data #ai #agents #theravitshow

𝐂𝐚𝐧 𝐘𝐨𝐮 𝐁𝐮𝐢𝐥𝐝 𝐚 𝐏𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧-𝐑𝐞𝐚𝐝𝐲 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭 𝐟𝐨𝐫 $𝟎? A year ago this stack meant paid APIs at every layer. Today the open-source ecosystem covers all of it. 𝐇𝐞𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐟𝐫𝐞𝐞 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞, 𝐥𝐚𝐲𝐞𝐫 𝐛𝐲 𝐥𝐚𝐲𝐞𝐫. 1. Frontend: Routes user requests from web, mobile, or API to your backend. Next.js or Streamlit, deployed on Vercel's free tier. The entry point users actually touch. 2. Agent Orchestrator: The system's brain running end-to-end data flow. LangGraph or CrewAI. Coordinates everything downstream planning, retrieval, execution. 3. RAG Pipeline: Retrieves relevant context for grounding. Pull knowledge from sources like Notion. Store in ChromaDB. Run Qdrant locally as your vector database. The layer that stops your agent from hallucinating. 4. LLM Layer: Run models locally with zero API costs. Ollama serving open models like Llama 3 or Mistral. This is where "$0" really shows up. No per-token billing, no rate limits, no vendor dependency. 5. Tools (MCP): Connect the agent to the real world via Model Context Protocol. GitHub, Slack, databases, file systems. How the agent goes from thinking to doing. 6. Code Agent: Writes, debugs, and generates code. Claude Code CLI or Aider. The builder inside your builder. 7. Data Layer: Store application state and structured data. SQLite, DuckDB, or Supabase. Lightweight, free, production-capable. 8. Deployment: Ship anywhere for free. Docker, Cloudflare Workers, or Hugging Face Spaces. Wrapping everything: observability with Phoenix (self-hosted), Prometheus, and Grafana to monitor, log, and improve. What's the key architectural decision? "Need more knowledge?" If yes, route through the RAG pipeline for context. If no, go straight to the LLM. That branch keeps the system efficient instead of running expensive retrieval on every request. What's the real takeaway? The barrier to building a serious AI agent isn't cost anymore. Orchestration, local inference, vector storage, tooling, observability, all covered by open source. The barrier is knowing how to wire the pieces together. Which layer would you swap out or upgrade first? ♻️ Repost this to help your network get started ➕ Follow Sathish for more #AIAgents #OpenSource #AIEngineering

Open-source AI isn’t just one category. It’s an ecosystem. Most people miss this. That’s why they struggle to pick the right tools. Here’s the clarity you actually need ⤵️ ✦ Open-Source Base Models Where everything begins. They give you text, image, speech, and multimodal power, with the freedom to fine-tune for your domain. Best for teams that want flexibility instead of lock-in. ✦ Model Deployment Tools This is the bridge from “cool demo” to “real product.” Scalable infra, production-ready APIs, private inference, everything enterprises need to operationalize AI. ✦ Specialized Libraries Your precision tools. Image cleanup, audio enhancement, extraction, conversions, targeted solutions that save you from building basics from scratch. ✦ RAG Engines Where models become knowledge-aware. They fetch, reason, and respond using your data, without retraining. Perfect for intelligence, documentation, and enterprise knowledge systems. ✦ LLM Frameworks The orchestration layer. Memory, tools, agents, routing, the components that help you build full AI workflows instead of one-off prompts. ✦ Agentic Frameworks The advanced layer. Systems that can plan, decide, execute, and complete multi-step tasks. This is where AI stops being a tool… and starts becoming a teammate. --- When you understand these layers, you stop “trying tools.” You start architecting solutions. And that’s the difference between experiments and real AI products. ____________________________ 📌 If you want a high-res PDF: 1. Follow Sufyan Maan, M.Eng. 2. Like the post. 3. Repost to your network. 4. Subscribe to: sufyanmaan.substack.com

🚨 The OSPO is evolving. Is yours keeping up? /* Most likely the last post on this topic in a while */ As organizations mature their oss strategies, OSPOs can no longer limit their value to compliance, contribution facilitation, and related policies, processes, and tooling. That was OSPO 1.0. It's 2025. Welcome to OSPO 2.0! The next-gen OSPO plays a much broader, more strategic role in helping organizations navigate the complexities of AI, ecosystem orchestration, developer enablement, and open innovation at scale. Here are some forward-looking areas OSPO leaders & executive sponsors should start planning for: ✅ Model governance for AI/ML: Help teams release models responsibly with the right licensing, transparency, and documentation. ✅ Open ecosystem strategy: Move from contributing to orchestrating ecosystems, including foundations, community governance, and standards participation. ✅ Developer enablement at scale: Become a partner to the platform and DevEx teams. Offer paved paths, internal OSS catalogs, and compliance-by-default tooling. Enable shift-left. ✅ Risk intelligence: Go beyond license compliance. Build signals into engineering workflows about project health, community maturity, and geopolitical risk. ✅ Global policy engagement: Interpret emerging regulations like the EU CRA/ AI Act/DMA , OSS mandates in China (Xinchuang/PIPL/DSL), US's Executive Orders (EO 14028), and advise on their impact on your strategy & products. ✅ Metrics & evidence-based or evidence-driven strategy: Use CHAOSS or other tools to create internal dashboards measuring OSS impact, maturity, and sustainability. 🎯 Next steps for OSPOs wanting to evolve: To be practical, here are a few suggestions on steps to take to start shifting from OSPO 1.0 to 2.0: ☑️ Set up an organizational cross-functional WG on open source and AI, figure out your position concerning Open Source Initiative (OSI)'s Open Source AI Definition (review & provide feedback), and the newly released The Linux Foundation OpenMDW 1.0 license (related post: https://lnkd.in/eg8GY8Y9). ☑️ Audit your current OSS footprint for gaps in tooling, model transparency, and project health ☑️ Map open source to your platform engineering strategy (where's the value line, where should you max your contributions) ☑️ Build policy awareness into OSPO reviews: AI, SBOM, regional compliance ☑️ Be on top of any updates to or new policies/regulations and how they affect your open source efforts OSPO 2.0 is a strategic enabler of trust, transparency, and technical leverage in a world increasingly shaped by open AI and software ecosystems. 📩 If you’re exploring how to evolve your OSPO or build one with these priorities, I’d be happy to connect. 👉 If this resonates, feel free to share it with others in your network. #OSPO #OpenSource TODO (OSPO) Group The Linux Foundation Linux Foundation Europe Linux Foundation Japan OpenChain Project SPDX SBOM CHAOSS Shuchi Sharma - following on our convo re: OSPO 2.0

💡 It’s been over 10 years since I tried to help a country in #LatinAmerica implement #CKAN through an open tender — only to discover that no consulting firm had the expertise to do it. Too often, open-source solutions are expected to be implemented by an “in-house team” that, in practice, lacks the mandate or resources to sustain them. So moving from “I have a proprietary solution and I am the only one to implement it” to “I have an open-source solution and you are the only one who can implement it” is real progress — but still not enough. What we really need is an active #TechEcosystem of reliable implementation partners. CKAN now has many, but for other #DPGs this is still missing. And nothing accelerates this more than: ✅ Great documentation ✅ Intentional ecosystem-building ✅ Well-designed tenders That’s how we’ll make open source truly sustainable.

Building momentum in any ecosystem—whether it's a business network, community, or partner ecosystem—requires both strategic alignment and practical activation. Most importantly, it needs someone to drive as the organizer/activator. Here's my framework for creating lasting momentum: 1. Clarify the Shared Purpose  Momentum comes from a compelling vision that stakeholders can see themselves as a part of. Define the collective "why"—what outsized impact can you create together that no single participant could achieve alone? 2. Start Small, Show Proof Don't try to boil the ocean. Launch small, visible projects that demonstrate value. Early wins attract more participants and build confidence in the ecosystem's potential. 3. Build Trust and Reciprocity Ecosystems thrive on mutual benefit. Create opportunities for knowledge sharing, co-marketing, and joint projects. Trust builds momentum faster than transactions—partnership = trust + true collaboration. 4. Enable Network Effects Structure so each new participant adds disproportionate value. Don't demand the same level of participation from everyone—tailor approaches to individual skills. The method of organization matters: How do people communicate, share resources, and ask questions? 5. Orchestrate, Don't Control This is such a unique and critical role. Provide lightweight governance—guide, convene, facilitate connections—but leave room for organic growth. Let people self-identify where they see themselves in the ecosystem. 6. Create Visibility and Buzz Celebrate milestones, share collaboration stories, and spotlight members. Momentum is social—people want to be part of something others are talking about. Communication is key. 7. Sustain Through Value Loops Participants give because they see value returned. Continuously measure and communicate the value each group receives, including leads, cost savings, learning opportunities, and social impact. The result? A flywheel effect where clarity of purpose, quick wins, trust, network effects, and continuous value loops create lasting ecosystem momentum. #EcosystemBuilding #CommunityBuilding #Leadership