Poolside

Poolside

Poolside trains open-weight mixture-of-experts models for software engineering and ships them through a terminal agent called pool, a macOS desktop app, IDE extensions, and an OpenAI-compatible inference API. Laguna S 2.1 and Laguna XS 2.1 are built in-house with reinforcement learning inside Poolside's agent harness, then released with downloadable weights rather than locked behind a hosted endpoint alone.

Most coding assistants assume a cloud API and per-seat billing. Poolside focuses on teams that need full model weights on their own hardware, air-gapped installs for defense workloads, and Agent Client Protocol hooks into VS Code, Zed, or JetBrains without replacing editors. Government deployments run unlimited on-premises inference with no per-token fees, which is a different cost model from token-metered copilots.

Platform engineers and enterprise dev teams use Poolside for long-horizon agentic coding, from Terminal-Bench style refactors to secure automation inside a VPC. Defense contractors and regulated shops pick it when a generic SaaS coding bot cannot cross the network boundary.

Top Features:
  1. Laguna S 2.1 MoE model with 118B total parameters, 8B active, and a 1M token context window

  2. Laguna XS 2.1 fits on-device at 33B total parameters with 3B active per forward pass

  3. pool CLI installs with curl -fsSL https://downloads.poolside.ai/pool/install.sh and supports ACP editors

  4. OpenAI-compatible API at inference.poolside.ai/v1 for poolside/laguna-s-2.1 and poolside/laguna-xs-2.1

  5. Government stacks deploy air-gapped with STIG-hardened options and no per-token inference fees

  6. Laguna models trained with RL inside pool on up to 30T tokens for S 2.1 and 15T tokens for XS 2.1

Pros:
  1. Laguna models ship as open-weight releases under permissive licenses from OpenMDW to Apache 2.0.

  2. On-premises and air-gapped deployment delivers full model weights, not API-only access.

  3. pool CLI and Agent Client Protocol integrations fit VS Code, Zed, JetBrains, and Visual Studio workflows.

  4. Government customers run unlimited on-premises inference with no per-token fees on their own hardware.

Cons:
  1. Enterprise deployment requires a sales conversation with no self-serve pricing page.

  2. Preview free access is described on the models page as available for a limited time only.

  3. The Poolside Desktop Assistant download is macOS-only on the get-started page.

  4. The full stack targets enterprises and regulated environments rather than casual hobby projects.

FAQs:

Does Poolside have public pricing?

Poolside does not publish self-serve enterprise pricing on its website. The models page lists Laguna S 2.1 and Laguna XS 2.1 as free for a limited preview window, while full on-premises and government deployments are sold through sales conversations.

Can I try Poolside models for free?

Yes. Poolside offers Laguna S 2.1 and Laguna XS 2.1 free in Preview through the pool CLI, its OpenAI-compatible API, OpenRouter, and the Vercel AI Gateway. The models page describes this preview access as free for a limited time.

What models does Poolside offer?

Poolside ships Laguna S 2.1, a 118B-parameter MoE model with 8B active parameters and 1M context, and Laguna XS 2.1, a 33B-parameter MoE model with 3B active parameters and 256K context. Both are trained in-house and released under permissive licenses from OpenMDW to Apache 2.0.

Where can Poolside be deployed?

Poolside supports on-premises, VPC, air-gapped, and disconnected environments for government and enterprise customers. The government page lists STIG-hardened OS options, IL5 deployability, and cleared resources for federal missions.

Which IDEs and editors does Poolside support?

Poolside connects to VS Code, Visual Studio, Zed, and JetBrains through the Agent Client Protocol. The get-started page also documents pool acp setup commands for Zed and JetBrains, plus a macOS desktop client.

Is Poolside API compatible with OpenAI?

Yes. Poolside exposes an OpenAI Chat-compatible API at inference.poolside.ai/v1, so teams can point existing OpenAI SDK clients at Poolside endpoints using models like poolside/laguna-s-2.1 without rewriting integration code.

How do I get Laguna model weights?

Poolside publishes Laguna family weights under permissive licenses from OpenMDW to Apache 2.0. The models page links to weight downloads and API keys for Laguna S 2.1 and Laguna XS 2.1, and invites the community to evaluate, fine-tune, and build on the releases.

Pricing:

Freemium

Tags:

Foundation Models
Coding Agents
Open Weight Models
MoE Models
On-Premises Deployment
Agent Client Protocol
Software Engineering
Enterprise AI

Tech used:

Svelte
Tailwind CSS
Vercel
Vercel Analytics
Ruby

Reviews:

Give your opinion on Poolside :-

Overall rating

Join thousands of AI enthusiasts in the World of AI!

Best Free Poolside Alternatives (and Paid)

By Rishit