Enterprise
Custom
On-prem, VPC, or cloud deployment
Poolside
Enterprise AI for software engineering — built on production code, built for production systems.
Poolside is an enterprise and government-grade AI software development platform with a proprietary model family trained on production code. The 2026 lineup includes Malibu (complex engineering tasks), Point (rapid code completion), and Laguna (open-weight, high-security deployments). Available via API, Amazon Bedrock, IDE plugins, and the new Poolside Console with plan mode, repositories, and third-party model support.
+ Pros
- Custom foundation models (Laguna XS.2 and M.1) trained entirely in-house from scratch with their own data pipeline, infrastructure, and async on-policy reinforcement learning — not a wrapper on top of existing APIs.
- Enterprise-grade security and deployment flexibility: deploy on-premises, air-gapped, in your VPC on AWS/Azure/GCP, or via turnkey hardware partners like Dell — data never leaves your control.
- Open-weight model (Laguna XS.2, 33B-A3B) released under Apache 2.0 license on Hugging Face, enabling community fine-tuning, quantization, and local serving on a single GPU.
- Competitive benchmark performance: Laguna M.1 (225B-A23B) achieves 65.4% on SWE-bench Verified, 57.4% on SWE-bench Multilingual, and 32.7% on Terminal-Bench 2.0 — competitive with models many times its size.
- Full auditability and governance with the Poolside Console: every agent action, tool call, and reasoning step is recorded as a searchable trajectory; role-based access control, redaction patterns, and audit logs are built-in.
− Cons
- Text-only models — Laguna XS.2 and M.1 do not support vision or multimodal inputs, limiting use cases that involve screenshots, diagrams, or UI analysis.
- Enterprise-focused with no public pricing — requires contacting sales for access; no self-serve signup or pay-as-you-go tiers available for individual developers.
- Relatively young ecosystem compared to incumbents — fewer community resources, third-party integrations, and proven track record versus OpenAI, Anthropic, or Google.
Pricing
AWS Bedrock
Pay-per-token
Managed endpoints via Amazon Bedrock
Poolside is an enterprise-focused AI software development platform that trains its own proprietary models on production-grade code. Founded in 2023, it raised a $2B round in late 2025 at a valuation of $12–14B with Nvidia as anchor investor, positioning itself as the high-security, high-compliance alternative to general-purpose coding assistants. The platform is used by large enterprises and government organizations that require data isolation, on-premise deployment, and full auditability.
The 2026 model family covers three tiers: Malibu handles complex, multi-step engineering tasks and agentic workflows; Point delivers rapid inline code completion; and Laguna — launched in April 2026 as the first open-weight models from Poolside — targets high-security environments such as government and regulated industries where self-hosted, auditable models are a requirement. The Malibu agent framework enables autonomous development workflows across planning, implementation, testing, and deployment stages.
In December 2024, Poolside announced a strategic partnership with AWS, making it the first cloud provider to offer Poolside models as managed endpoints on Amazon Bedrock. The Poolside Console — launched in April 2026 — added a developer-facing interface with plan mode, repository management, and third-party model provider support, giving enterprises a unified control plane for their AI engineering workflows. Poolside is also pursuing Project Horizon, a planned 2GW data center in Texas to support sovereign AI infrastructure at scale.
Key Features
• Malibu: flagship model for complex, multi-step agentic engineering tasks
• Point: fast inline code completion model for IDE integration
• Laguna / Laguna XS.2: open-weight models for air-gapped and high-security deployments
• Malibu agent framework: autonomous workflows across the full SDLC
• Poolside Console: plan mode, repository management, third-party model support (April 2026)
• Amazon Bedrock: managed Poolside endpoints, pay-per-token, VPC-compatible
• IDE plugins: VS Code and JetBrains integration via API
• Enterprise compliance: data isolation, on-premise/VPC deployment, SOC 2, audit logs
• Government-grade: Laguna open-weight for sovereign, air-gapped environments
• $2B raised · $12–14B valuation · Nvidia anchor investor · Project Horizon (2GW Texas)
Version History
Both models upgraded to 256K context window. Laguna M.1 reaches 45.8% on Terminal-Bench 2.0. Over 1 trillion tokens processed and 50,000+ XS.2 downloads in 4 weeks.
Production-grade enterprise platform for running AI agents inside your security boundary. Includes Console with agent management, sandboxed execution, audit trails, MCP server integration.
Terminal-based coding agent with interactive mode, ACP server/client support, sandboxed execution, MCP integration, and non-interactive 'pool exec' for CI/CD.
Two foundation models for agentic coding: Laguna M.1 (225B-A23B) and Laguna XS.2 (33B-A3B, Apache 2.0). Launched alongside pool CLI and Shimmer cloud dev environment. Trained on 30T tokens using 6,144 NVIDIA H200 GPUs.
Added Skills, Credentials/Secrets, Audit Log, Redaction Patterns, agent trajectory export, plan mode (beta), third-party model providers (beta), and OpenShift/Kubernetes deployment support.
Signature Snippet
An enterprise team deploys Poolside Malibu via Amazon Bedrock inside their VPC: the model reads a 2M-line Java monolith, generates a migration plan to microservices, writes and tests the decomposed services against the company's CI standards, and submits PRs following internal code conventions — with zero data leaving the secure perimeter.More in this Space
Vix
Vix is a Go-native, open-source (AGPL-3.0) AI coding agent that slashes token costs by 40-50% using a stem agent architecture and Tree-sitter virtual filesystem. It rethinks the plan/execute loop — keeping LLM cache warm across Explore, Plan, and Execute phases — while shipping Programmable Workflows, Whiteboard Mode with voice AI, MCP server support, and a self-evolving agent that writes its own scheduled jobs and watchers.
Late — High-Leverage AI Agent Orchestration
Orchestrate an entire AI dev team on 5GB VRAM using ephemeral subagents, exact-match diffs, and a zero-dependency Go binary. Works with any OpenAI-compatible model — local or cloud.
Paca
Paca is a free, open-source, self-hosted Scrum board where AI agents work as equal teammates — assigned to sprints, picking up tasks, and collaborating on BDD specs alongside humans. Built as an alternative to Jira and Linear, it treats AI agents as first-class Scrum members.