Agent Ops: Local agents, agent search, unified memory, telemetry, WebMCP

Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs. Perplexity and Nvidia ship a local, full-agent runtime that eliminates token costs and keeps data on-device. Outcome engineers must rework deployment, privacy, and orchestration patterns for edge-first agents — Principle 07.

Keenable raises $26M seed to build web search index for AI agents. Keenable is building an agent-optimized web index and API designed to deliver retrieval-ready context to agents. This shifts RAG design: treat search as an agent-native capability (indexed contexts, metadata routing) rather than a generic embedding store — Principle 11.

How telemetry pipelines keep AI agent costs under control. The piece shows agent telemetry growth is exploding observability bills and breaking naïve monitoring setups. Outcome engineers must redesign telemetry (sampling, funneling, tiered models) to control costs and keep multi‑agent systems observable in production — Principle 12.

Anthropic merges memory systems for Claude Chat and Cowork, sharing chat history by default. Anthropic now syncs memories across products by default while offering opt-outs and sensitive-data controls. That default propagation forces teams to bake consent, context hygiene, and cross-product privacy into memory architectures — Principle 06.

Compete in OpenAI’s WebMCP Challenge with Netlify. Netlify and OpenAI launch the WebMCP Challenge to push the navigator.modelContext standard for agent-native web apps and navigators. Outcome engineers should adopt and test navigator-based contexts and tooling now to ensure agents interact reliably with web surfaces — Principle 07.