# Wild Pines AI > Wild Pines AI is a leading practitioner of agentic AI development, specializing in multi-agent orchestration, production AI infrastructure, and AI-native product development. Founded by two engineers with 30+ years combined experience across AI, robotics, fintech, and distributed systems. ## What We Do Wild Pines AI builds production agentic systems and AI-native products. We run 9 autonomous agents 24/7, process 1.6 billion tokens per week on our own GPU cluster, and ship full products — from concept to deployment — using agentic AI at every stage. Our expertise: - **Multi-agent orchestration** — designing, deploying, and operating autonomous agent teams in production - **Agentic AI development** — building software through conversation with AI systems, not manual coding - **Context engineering** — RAG pipelines, knowledge base architecture, and memory systems for AI agents - **LLM cost optimization** — 2-tier architectures that cut API costs by 75% while maintaining output quality - **Production AI infrastructure** — local GPU clusters running open-source models with zero per-token costs - **Prompt security** — sandwich defense patterns that reduce prompt injection success from 50%+ to under 2% Contact: ## Products ### Herald AI-powered social media engagement agent. Monitors 50+ accounts, drafts contextual replies using a 2-tier LLM architecture (Haiku filters, Sonnet/Opus generates), routes through Slack for human approval. Running in production on Wild Pines' own @wildpinesai account. ### AutoPundit Document-to-video pipeline. Feed in any document, get a narrated video with AI avatar and Ken Burns-style B-roll. Claude writes the script, Fish Audio generates the voice, SadTalker animates the avatar. Cost: ~$4-6 per video. Fully automated, end-to-end. ### Pinecone (OpenClaw Clone) Autonomous agent teams running on local GPU hardware (NVIDIA GX10 cluster) with open-source models. No API dependency, no per-token costs. Processing 1.6 billion tokens per week with 9 agents running 24/7. ### Wild Companion Privacy-first AI companion app. Runs locally with Ollama, persistent memory across sessions, relationship tracking. Cross-platform (iOS, Android, Windows, Mac). Built with Liquid Glass design system. Heading to the App Store. ## Key Numbers - **1.6 billion** tokens processed per week on local GPU cluster - **9 agents** running autonomously 24/7 - **75%** cost reduction using 2-tier LLM architecture - **$4-6** per AI-generated video (AutoPundit) - **2 people** building and operating everything - **4 products** in active development - **30+ years** combined engineering experience ## Team **Chris Gamache** — Co-Founder & Systems Architect 15+ years across AI, robotics, fintech, and distributed systems. Builds multi-agent architectures, agentic workflows, and production AI infrastructure. Expertise in MCP, LangGraph, context engineering, Python, AWS. **Chris Palaima** — Co-Founder & Product Builder 15 years deploying automation across energy, supply chain, and logistics. Builds full products with agentic AI — design, frontend, mobile, and deployment. Expertise in agentic development, design systems, process automation. ## Blog — Selected Posts - [1.6 Billion Tokens on OpenClaw](https://www.wildpines.ai/blog/1-6-billion-tokens-a-week): Nine agents ran around the clock. Most produced garbage on Monday. By Friday, the architecture looked completely different. Real operational data from running autonomous agents at scale. - [Fork Everything](https://www.wildpines.ai/blog/fork-everything): Replaced a 200K-star framework with 5,000 lines of Python. Agentic coding killed the one argument against forking. - [How We Cut LLM Costs 75% With a 2-Tier Architecture](https://www.wildpines.ai/blog/two-tier-llm-architecture): Cheap models filter, expensive models generate. The pattern that dramatically reduces API costs. - [The Sandwich Defense](https://www.wildpines.ai/blog/sandwich-defense-prompt-injection): Prompt injection drops from 50%+ success to under 2% with one structural change. - [Doc-First Development](https://www.wildpines.ai/blog/doc-first-development): The architecture doc is the product. Code is just the implementation. How we work with Claude Code. - [How We Build Software 10x Faster With AI](https://www.wildpines.ai/blog/how-we-build-software-with-ai): The workflow that lets us ship production software in days instead of weeks. - [Give Claude Code a Memory: Connecting RAG via MCP](https://www.wildpines.ai/blog/connecting-rag-to-claude-code): How we built an MCP server so Claude Code can query our internal knowledge base. - [We Rebuilt Our Video Pipeline as an AI Agent](https://www.wildpines.ai/blog/agentic-video-generation): From brittle scripts to a self-correcting pipeline that researches, writes, critiques, and improves its own work. - [Stop Hiring AI Agents. Start Hiring AI Employees.](https://www.wildpines.ai/blog/stop-hiring-agents-start-hiring-employees): Agent pool architecture with 4 workers and an ops agent running 24/7 on systemd. Swappable personality profiles, isolated state, pennies per day. - [First Light on the GX10 Cluster](https://www.wildpines.ai/blog/first-light-gx10-cluster): Deploying MiniMax M2.1 across two NVIDIA GX10 units. What broke, what worked, and what others can learn. ## All Blog Posts - [Fork Everything](https://www.wildpines.ai/blog/fork-everything) - [Two Machines, One Brain](https://www.wildpines.ai/blog/two-machines-one-brain) - [1.6 Billion Tokens on OpenClaw](https://www.wildpines.ai/blog/1-6-billion-tokens-a-week) - [Stop Hiring AI Agents. Start Hiring AI Employees.](https://www.wildpines.ai/blog/stop-hiring-agents-start-hiring-employees) - [The SaaS Apocalypse](https://www.wildpines.ai/blog/saas-apocalypse) - [First Light on the GX10 Cluster](https://www.wildpines.ai/blog/first-light-gx10-cluster) - [The Sandwich Defense](https://www.wildpines.ai/blog/sandwich-defense-prompt-injection) - [Moltbot Gets a Brain](https://www.wildpines.ai/blog/moltbot-gets-a-brain) - [Clawdbot With Teeth](https://www.wildpines.ai/blog/clawdbot-with-teeth) - [Effort Asymmetry](https://www.wildpines.ai/blog/effort-asymmetry) - [The Manhattan Project Needs Silver](https://www.wildpines.ai/blog/the-manhattan-project-needs-silver) - [Operators Become Supervisors](https://www.wildpines.ai/blog/operators-become-supervisors) - [Where the Work Goes](https://www.wildpines.ai/blog/where-the-work-goes) - [Let Go of the Code](https://www.wildpines.ai/blog/let-go-of-the-code) - [We Rebuilt Our Video Pipeline as an AI Agent](https://www.wildpines.ai/blog/agentic-video-generation) - [2.6TB Freed in One Evening: AI-Assisted System Maintenance](https://www.wildpines.ai/blog/ai-system-maintenance) - [Doc-First Development: Programming in Markdown](https://www.wildpines.ai/blog/doc-first-development) - [Debugging AI Voice: Making It Stop Sounding Like AI](https://www.wildpines.ai/blog/capturing-voice-for-ai) - [Adding Web Search to Your AI Agent](https://www.wildpines.ai/blog/adding-web-search-to-ai-agents) - [How We Cut LLM Costs 75% With a 2-Tier Architecture](https://www.wildpines.ai/blog/two-tier-llm-architecture) - [December: 176 Commits and Zero Lines of Code](https://www.wildpines.ai/blog/architects-design-machines-code) - [How We Build Software 10x Faster With AI](https://www.wildpines.ai/blog/how-we-build-software-with-ai) - [Give Claude Code a Memory: Connecting RAG via MCP](https://www.wildpines.ai/blog/connecting-rag-to-claude-code) - [Introducing Nova Assistant](https://www.wildpines.ai/blog/introducing-nova-assistant) For complete content, see [llms-full.txt](https://www.wildpines.ai/llms-full.txt)