Vscode Extension vs Poolside AI

The feature lists barely intersect. Poolside ships models and a platform: Laguna XS 2.1 (33B params, 3B active, 256K context, built to run on-device) and Laguna S 2.1 (118B, 8B active, 1M context, aimed at long-horizon reasoning), plus single- and multi-agent orchestration with planning and tool use, sandboxed execution environments, data connectors to repositories, databases, and warehouses, and model selection, tool grouping, and project management in a desktop app and CLI. Its surfaces are OpenRouter and Vercel AI Gateway; its governance layer covers RBAC for both human users and agents, end-to-end trace observability, and auditability.

The VS Code Extension is an editor experience: context-aware completions, multi-file real-time edits, panel and inline chat, quick chat on a keyboard shortcut, voice chat, terminal chat, Smart Variables for referencing code elements in prompts, AI symbol renaming, automatic commit messages and PR descriptions, token-usage monitoring, and bring-your-own-model configuration. Integrations are Gemini, GitHub Models, Ollama, LMStudio, GitHub Copilot Extensions, and the VS Code Marketplace.

Capability depth differs by intent: Poolside's hard problems are orchestration, sandboxing, and proving what an agent did; the extension's hard problems are latency, in-editor ergonomics, and provider flexibility. One builds a governed agent runtime; the other wires your chosen model into a code editor.

Poolside's pricing_type is contact: there is no published per-seat tier, so no self-serve trial and no instant sign-up. Budget approval requires a sales conversation, which is exactly the friction its 'not_for' list acknowledges. Whatever the contract lands at, the cost includes the models themselves — open weights you can inspect, fine-tune, and host, so infrastructure spend replaces per-token billing. The 2026-08-22 Bloomberg report of a $6B Nvidia licensing payment is a signal about Poolside's backing and commercial direction, not a price you can plan against; treat any number here as enterprise-negotiated. Factor in GPU capacity if you self-host Laguna S 2.1's 118B parameters, and far less for the 3B-active XS 2.1 on-device.

The VS Code Extension is free — the IDE costs nothing, and your only cost is what your model provider charges on your own API key. Teams using free provider tiers can push AI tooling costs near zero, which is precisely the persona Flexpilot targets. The trade is unbundled: no managed billing, no admin console, no support SLA, and you own quota management and key rotation. So the two pricing models answer different questions. Poolside: 'what does a governed, self-hosted agent platform cost us?' The extension: 'how little can we spend while keeping control of the keys?'