Poolside
Open-weight agentic coding models — Laguna XS 2.1 at 33B on-device and Laguna S 2.1 at 118B for long-horizon agent work.
Pick Poolside when the constraint isn't your editor but your security boundary — the open weights plus on-prem, VPC, and air-gapped options are the actual product. Laguna S 2.1 (118B/8B active/1M context) and Laguna XS 2.1 (33B/3B active/256K, on-device capable) give you a real size range without leaving your perimeter. If you just want faster autocomplete for ten engineers, GitHub Copilot or Cursor will cost less and ship sooner; Poolside is a joint engineering engagement, not a seat license.
Verified 9h ago · liveness 68/100 · cite: rightaichoice.com/tools/poolside
- Regulated enterprises in defense, finance, or healthcare that need on-prem or air-gapped agentic coding
- Platform and DevEx teams required to keep code and agent execution inside their own security boundary
- CISOs and compliance leads who need RBAC plus end-to-end audit trails on coding agents
- Engineering orgs with large monorepos and long-horizon tasks that break autocomplete tools
- Individual developers who want a cheap coding assistant today
- Startups and small teams without a procurement cycle or compliance mandate
- Teams doing simple, repetitive coding work that doesn't need multi-agent planning
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Poolside if you want a low-cost coding assistant you can adopt without a procurement cycle or a vendor engineering engagement.
Self-hosting open-weight models moves inference onto your own GPUs, so the hardware and ops cost lands on your infrastructure budget rather than a subscription line.
Poolside is structured for enterprise and public-sector procurement rather than self-serve seats, which places it against GitHub Copilot and Cursor at the low end and against other privately deployed agentic coding stacks at the high end. The open-weight Laguna models, including the 33B/3B-active Laguna XS 2.1 that runs on-device, are the escape hatch for teams that would rather pay in infrastructure than in licenses.
In short
Poolside — Open-weight agentic coding models — Laguna XS 2.1 at 33B on-device and Laguna S 2.1 at 118B for long-horizon agent work. Best for Regulated enterprises in defense, finance, or healthcare that need on-prem or air-gapped agentic coding, Platform and DevEx teams required to keep code and agent execution inside their own security boundary, CISOs and compliance leads who need RBAC plus end-to-end audit trails on coding agents. Contact Sales pricing.
No head-to-head comparison for Poolside yet — the alternatives below are the closest matches we list.
What's new in Poolside
Checked 3 days agoAcross the latest 3 updates: 2 launches and 1 changelog entry.
Introducing Laguna S 2.1
Poolside released Laguna S 2.1, a model upgrade aimed at longer-horizon work and heavier reasoning use.
Introducing Laguna XS 2.1
Laguna XS 2.1 ships as an upgraded version of the Laguna XS.2 model.
Long context update: Laguna XS.2 and M.1
Laguna XS.2 and M.1 now served at 256K context lengths.
What people actually say about Poolside — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
87 mentions across 5 sources (Hacker News, YouTube, Product Hunt, App Store, Lemmy) · researched Aug 24, 2026.
Average across the 5 sources that answered — each source counts once, not each post.
- +Open-weight models (Apache 2.0) allow full self-hosting and fine-tuning.
- +Long context windows (256K/1M) handle entire codebases per query.
- +Strong non-coding intelligence was noted by one HN user, praising its versatility.
- +Multi-agent orchestration with planning and tool use supports complex tasks.
- +On-prem, VPC, or air-gapped deployment meets strict regulatory requirements.
- −Local inference (llama.cpp) can corrupt context within 10k tokens (Q4).
- −Early releases had serious bugs; fixes came quickly but user patience is tested.
- −Pricing is unclear (contact sales), likely beyond budget for startups.
- −Ease of use is low; requires significant setup and enterprise buy-in.
- −Not a top choice for general coding; DeepSeek and others outperform for many.
- • Potential infrastructure costs for self-hosting GPUs
- • Professional services fees for custom deployments
- • No transparent public pricing; likely requires a sales engagement
Viability Score
How well maintained and how widely used is Poolside? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Open-weight agentic coding models: Laguna S 2.1 and Laguna XS 2.1
- Laguna S 2.1: 118B params, 8B active, 1M context, released 2026-07-21 for long-horizon agent work
- Laguna XS 2.1: 33B params, 3B active, 256K context, small enough to run on-device
- Laguna XS.2 and M.1 served at 256K context after the 2026-05-26 long-context update
- Agent runtime with planning, tool use, and sandboxed code execution
- Poolside Platform deploys agents inside your own security boundary
- RBAC and end-to-end audit trails for agent governance
- On-prem, VPC, and air-gapped deployment with self-hosted model download
- Desktop app for agentic coding sessions
- pool CLI v1.0.6: projects, connectors, plan and agent modes, slash commands
- Models served via OpenRouter and Vercel AI Gateway
- Forward Deployed Research Engineers embedded with customer teams
- Custom evaluations co-created against your own codebase
- Dedicated government track for public-sector missions
About Poolside
Poolside builds open-weight foundation models for agentic software engineering, plus the runtime that drives them. Two models are current: Laguna S 2.1, the newest release (2026-07-21) at 118B params / 8B active / 1M context, tuned for long-horizon agent work and heavier reasoning at mid-size cost; and Laguna XS 2.1 (2026-07-02), a 33B param / 3B active / 256K context upgrade to XS.2 that is small enough to run on-device. The Laguna family started with XS.2 and M.1 in March 2026; a 2026-05-26 update moved both to 256K serving context. This is agentic coding rather than autocomplete. The models carry planning, tool use, and sandboxed execution, and you drive them through a desktop app, the pool CLI (v1.0.6, with projects, connectors, plan and agent modes, and slash commands), or hosted endpoints on OpenRouter and Vercel AI Gateway. Who it's for: engineering organizations that cannot route code through someone else's cloud. The Poolside Platform runs agents inside your own security boundary with RBAC and end-to-end audit trails, and supports on-prem, VPC, and air-gapped deployments with self-hosted model downloads. Deployments pair with Forward Deployed Research Engineers who co-build custom evaluations against your codebase, and there's a separate government track for public-sector missions. Positioning: Poolside is not competing with seat-priced coding assistants on price per developer. It sells a model-plus-runtime stack plus joint engineering, and it publishes its research openly, including the six-part Model Factory series on training and evaluation. That makes it a fit for teams whose bottleneck is monorepo-scale, long-horizon work under a compliance regime — not for a solo developer who wants a cheap tab-complete today.
Behind the Verdict
The pitch that matters here isn't model quality, it's deployment control. Most agentic coding tools assume your source code can leave your network. Poolside assumes the opposite: open-weight models you can download and run, a runtime that operates inside your own boundary, RBAC, audit trails, and an air-gapped option. For a defense contractor, a bank, or a hospital system with a hard rule about where code lives, that changes the procurement conversation more than a benchmark score does.We'd reach for this when the work itself is long-horizon. The 1M context on Laguna S 2.1 is aimed squarely at agent sessions that span a large repo and many tool calls, and the 33B/3B-active XS 2.1 gives you something you can actually run locally when latency or data residency rules out a hosted endpoint. Pairing those with Forward Deployed Research Engineers and evaluations co-built against your own codebase is the part most vendors won't do — it's a joint engineering commitment, not a support ticket.Where it bites: adoption is heavier than a plugin install. Expect a procurement cycle, a deployment project, and vendor engineering involvement before you see value. Teams that just want an assistant in the IDE will find this stack substantially more machinery than the problem requires.Closest alternative: GitHub Copilot or Cursor for individual velocity, and self-hosted open-weight coding models if you have the platform team to run inference yourself. Poolside sits between those — open weights and self-hosting, but with the agent runtime, governed deployment, and research engineers attached. If you already run
Researching Poolside? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Poolside actually fits — and what changes day-one when you adopt it.
You need agentic coding on a codebase that cannot leave an air-gapped network. Download Laguna S 2.1 (118B params, 8B active) into your own boundary, run multi-agent refactor jobs through the Poolside Platform with sandboxed execution, and keep RBAC plus end-to-end audit trails on every change.
Outcome: Agent-driven refactoring runs entirely inside your perimeter, with an audit record your compliance lead can hand to an assessor.
You want to see whether agentic coding fits your workflow before anyone talks procurement. Run Laguna XS 2.1 (33B params, 3B active, 256K context) on-device through the desktop app or the pool CLI, and use plan mode to scope a migration before letting an agent execute.
Outcome: You get a working read on model quality against your own code in a day, with no data leaving your machine.
Try the models hosted on OpenRouter or Vercel AI Gateway against a real branch, then bring in Forward Deployed Research Engineers to co-build evaluations on your repository and decide whether to move to a VPC deployment.
Outcome: The build-versus-buy decision is made against measured results on your codebase rather than a generic benchmark.
Use Cases
- Run agentic coding inside your VPC or air-gapped network where code cannot leave the perimeter.
- Orchestrate multiple agents to plan and execute migrations across legacy systems.
- Process a large monorepo context in one pass using Laguna S 2.1's 1M context window.
- Run the lightest agentic coding model on a laptop with Laguna XS 2.1 at 33B params / 3B active.
- Keep RBAC and end-to-end audit trails on every AI-assisted code change for regulated work.
- Co-build custom evaluations with Poolside research engineers against your own codebase.
- Prototype on OpenRouter or Vercel AI Gateway before committing to a private deployment.
- Bring open-weight models to public-sector missions through Poolside's government track.
Models Under the Hood
as of 2026-10-16
Limitations
- Poolside sells a deployment, not a download-and-go tool.
- The models are open-weight and self-hostable, but production use implies real infrastructure: Laguna S 2.1 carries 118B params (8B active) and Laguna XS 2.1 carries 33B (3B active), and the Platform runs agents on-prem, in a VPC, or air-gapped — all of which you have to stand up and operate.
- Agentic work is embedded in an engagement model: Forward Deployed Research Engineers co-build your evaluations, so adoption is a joint effort with a vendor team rather than a self-serve signup.
- The CLI is at v1.0.6 and the public surface is desktop, CLI, and hosted gateways, so the product is early relative to the model research.
- On research claims, the published C2C activation offloading result is a training-throughput finding (6-13% over selective activation checkpointing on Grace Blackwell), not a user-facing performance guarantee.
as of 2026-10-01
Verification history
We have re-verified Poolside 20 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 20 verification passes.
Free to cite with attribution — this page re-verifies continuously.
Where the pricing makes sense
The company stage and team size where Poolside's pricing actually pencils out — and where peers do it cheaper.
Poolside is structured for enterprise and public-sector procurement rather than self-serve seats, which places it against GitHub Copilot and Cursor at the low end and against other privately deployed agentic coding stacks at the high end. The open-weight Laguna models, including the 33B/3B-active Laguna XS 2.1 that runs on-device, are the escape hatch for teams that would rather pay in infrastructure than in licenses.
Setup time & first value
How long it actually takes to get something useful out of Poolside — broken out by persona, not the marketing-page minute.
On-device with Laguna XS 2.1 via the desktop app or pool CLI: same day, since a 33B/3B-active model runs locally. Hosted evaluation on OpenRouter or Vercel AI Gateway: same day, no infrastructure. VPC or on-prem deployment of Laguna S 2.1 with the Poolside Platform and custom evaluations: weeks, because it is scoped as a joint engagement with Forward Deployed Research Engineers. Air-gapped
Switching to or from Poolside
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From GitHub Copilot: move the agentic, multi-file work Copilot handles poorly onto Laguna models, keeping simple tab completion wherever it still fits.
- →From Cursor: lift refactors and migrations that outgrow a single editor session into multi-agent planning and sandboxed execution on the Poolside Platform.
- →From a hosted coding API: switch to open-weight Laguna models you download and serve yourself when code cannot leave your boundary.
- →From Laguna XS.2 or M.1: move to XS 2.1 or S 2.1 for the newer model generation, with 256K context still available on the earlier pair.
- ↗To GitHub Copilot or Cursor: if agentic planning and self-hosting turn out to be unnecessary, per-seat assistants cover day-to-day editing.
- ↗To a hosted coding API: if you decide code can leave your perimeter after all, a managed endpoint removes the GPU and ops burden.
- ↗To a smaller Laguna model: drop from S 2.1 to XS 2.1 on-device if latency or infrastructure cost matters more than long-horizon reasoning.
Integrations
Resources & Guides
- Resourcepoolside.ai
The latest stories from Poolside
Thoughts from the team building the foundation models and platform redefining how software gets made—securely, intelligently, and at scale.
- Resourcepoolside.ai
Upcoming events Poolside is attending
Meet our team and learn how to power your organization with foundational AI.
Tutorials & Learning
YouTube returned 6 videos for “Poolside”, and we withheld 6: 6 could not be judged, because “Poolside” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Poolside.
Official links
Tools that pair well with Poolside
Common stack mates teams adopt alongside Poolside, with the specific reason each pairing earns its keep.
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Kimi K
Kimi K2.6 brings open-weight long-horizon coding agents — 12-hour autonomous runs and 300-agent swarms.
Qwen3.6-35B-A3B
Open-weight Qwen3.6-35B-A3B runs agentic coding and reasoning locally with ~3B active parameters per token.
Alternatives to Poolside
View all
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Kimi K
Kimi K2.6 brings open-weight long-horizon coding agents — 12-hour autonomous runs and 300-agent swarms.
Qwen3.6-35B-A3B
Open-weight Qwen3.6-35B-A3B runs agentic coding and reasoning locally with ~3B active parameters per token.
Frequently Asked Questions
Best-of guides
Used Poolside? Help shape our editorial sentiment research.