Blaxel

Executive Summary

Blaxel calls itself the perpetual sandbox platform: microVM sandboxes for AI agents that automatically scale to zero when idle — at $0 compute cost — and resume in under 25 milliseconds with full memory state, filesystem, and running processes preserved.^[1][2] Where most sandbox vendors started ephemeral and bolted persistence on, Blaxel's architecture treats hibernation as the default lifecycle, the laptop-lid model: close it, it suspends; open it, it resumes where it left off.^[3] Around the sandboxes it layers a broader agent cloud — colocated serverless agent hosting, MCP server hosting, batch jobs, and a distributed filesystem ("Agent Drive").^[1]

The six-founder French-rooted team — whose previous data platform was acquired by OVHcloud — graduated Y Combinator's Spring 2025 batch and closed a $7.3M seed led by First Round Capital one month later, with YC, Liquid2, and Transpose participating.^[4][5] At the time of the raise the platform was serving 7.5M+ agent requests per day across 16 regions, with one customer running over 1 billion seconds of agent runtime; Webflow uses Blaxel sandboxes to power real-time previews for its AI coding agent.^[4] Named customers also include Shortwave, Strapi, Sapiom, Vybe, and Casco.^[1]

---

Product Overview

Blaxel's core loop: create a microVM sandbox, let the agent work in it, and stop thinking about teardown. After a short idle window the sandbox snapshots to standby — preserving filesystem state and running processes — and any incoming connection wakes it in under 25ms.^[2] The platform claims support for 50K+ concurrent sandboxes per workspace tier, scaling to 100,000+ on the top automatic tier.^[1][7]

The second half of the pitch is colocation: agent logic can be hosted serverlessly on the same infrastructure as the sandboxes it controls, eliminating the network round-trips that dominate latency when the agent runs on one cloud and its sandbox on another.^[3]

Key Capabilities

Product Surfaces

---

Technical Architecture

Each sandbox runs in an individual microVM with its root filesystem held in memory, which Blaxel pitches as both an isolation and a data-destruction guarantee — terminate the sandbox and the filesystem is gone.^[1] The specific hypervisor is not publicly disclosed; the docs describe "lightweight, sandboxed virtual machines" without naming Firecracker or an alternative.^[2] Standby snapshots preserve filesystem and process state but not external connections — database handles, message-queue consumers, and HTTP pools time out across hibernation, a real constraint for long-lived agent daemons.^[2]

Key Technical Details

---

Strengths

• Hibernation economics are the product — $0 compute on standby with sub-25ms resume means millions of stateful sandboxes can sit warm indefinitely, a model ephemeral-first competitors retrofit rather than start from.^[1][2]
• Verified production traction for a seed-stage company — 7.5M+ requests/day, billions of GB-seconds monthly, and a customer past 1 billion seconds of runtime, all reported at the July 2025 raise.^[4]
• A recognizable anchor customer — Webflow runs its AI coding agent's real-time previews on Blaxel sandboxes, exactly the wake-on-demand workload the architecture targets.^[4]
• More than a sandbox API — colocated agent hosting, MCP servers, and batch jobs make it a coherent agent cloud rather than a single primitive.^[1]
• Experienced repeat founders with strong backing — the team previously sold a data platform to OVHcloud, and First Round led the seed one month out of YC.^[3][4]

---

Cautions

• Managed-only, no BYOC or self-hosting — teams needing sandboxes inside their own VPC, GPUs, or a fuller in-VPC stack are explicitly outside the fit, per competitor analysis (Northflank — a rival, so discount accordingly).^[8]
• Hibernation drops external connections — database, queue, and HTTP-pool connections are not preserved across standby, so agents must be written to reconnect on wake.^[2]
• Hypervisor opacity — "microVM" is asserted but the isolation stack is not documented to the depth security-sensitive buyers will want; competitors like E2B publish their Firecracker architecture in the open.^[2][6]
• Small open-source footprint — the platform is closed and the MIT SDKs total a few dozen GitHub stars, so there is no community-auditable core and meaningful lock-in to a $7.3M-funded vendor.^[6][4]
• Heavy reliance on competitive content marketing — much of Blaxel's search presence is its own "E2B alternatives"-style SEO posts; independent third-party evaluation remains thin.^[9][10]
• Support costs extra, with a usage surcharge — email support is $800/month plus 3% of usage; live Slack support is $1,600/month plus 10%.^[7]

---

What Developers Say

There is no dedicated HN launch thread as of June 2026; Blaxel surfaces mainly in comparison comments and in one substantive production account — the maintainer of Broccoli, a Show HN cloud coding agent, describing a migration off Google Cloud Run.^[10]

"We also send about 50% of jobs to Blaxel; we find it has a much better cold start compared to Cloud Run." — yzhong94, Broccoli maintainer, on Hacker News^[10]

"We switched to Blaxel and it's much faster now. The hibernate feature has been great for comment iteration." — yzhong94 on Hacker News^[10]

"The closest equivalent is blaxel.ai, but it caters for AI SaaS startups, not individual customers." — leventov on Hacker News^[10]

"[Blaxel] falls short when you need BYOC, compliance, GPUs, or a full stack… in your own VPC." — Northflank's competitive analysis (a direct rival; read as adversarial)^[8]

One caveat on HN sentiment: co-founder Nicolas Lecomte participates in these threads (e.g., "E2B feels more ephemeral while Blaxel feels more stateful"), so some of the framing in comparison discussions is vendor voice.^[10]

---

Pricing & Licensing

No base subscription — everything is metered, with up to $200 in free credits and no credit card required to start.^[7]

Representative usage rates: sandbox runtime from $0.000023/s for 2GB (≈$0.08/hr) to $0.000368/s for 32GB (≈$1.32/hr); agent runtime $0.0000095/GB-s; MCP servers $0.000007/GB-s; batch jobs $0.000006/GB-s; volume storage $0.00000004629/GB-s; cron jobs free. All pricing as of June 2026.^[7]

Licensing model: Proprietary managed platform; SDKs, CLI toolkit, and agent-skills repos are MIT-licensed on GitHub.^[6]

Hidden costs: Standby sandboxes still accrue storage charges; paid support adds $800–$1,600/month plus a 3–10% usage surcharge; HIPAA compliance is a $250/month add-on.^[7][2]

---

Competitive Positioning

Direct Competitors

When to Choose Blaxel Over Alternatives

• Choose Blaxel when: sandboxes must hold user state across hours or days of inactivity without paying for idle compute, wake latency is user-facing (e.g., live previews), and you want agent logic hosted next to its sandbox.
• Choose E2B when: you want the most battle-tested option, an open-source core you can audit or self-host, and the deepest ecosystem.
• Choose Daytona when: creation speed, GPU sandboxes, Computer Use, or self-hosting drive the decision.
• Choose Sprites when: you are already on Fly.io and want stateful sandboxes inside that platform.

---

Ideal Customer Profile

Best fit:

• AI SaaS products giving each end user a long-lived workspace — coding-agent previews, data-analysis sessions, per-user dev environments — where idle time dwarfs active time
• Teams whose agent latency budget is dominated by sandbox cold starts or cross-cloud round-trips
• Startups that want sandboxes, agent hosting, and MCP servers from one usage-billed vendor with no platform fee

Poor fit:

• Organizations requiring BYOC, self-hosting, or an auditable open-source isolation layer
• GPU-dependent agent workloads
• Long-running daemons that must hold database or queue connections open continuously across idle periods

---

Viability Assessment

The traction-to-age ratio is the standout: billions of aggregate agent requests and 7.5M+ daily requests within months of founding is unusual, and First Round leading a $7.3M seed off the back of it is meaningful external validation.^[4] The structural risk is that persistence is converging across the category — every major sandbox vendor now offers some snapshot/resume story — leaving Blaxel to compete on the economics ($0 standby) and the colocated platform breadth.^[8]

---

Bottom Line

Blaxel is the strongest pure bet on the persistent-sandbox thesis: if your product hands every user an agent workspace that must survive days of idleness and wake instantly, its $0-standby, sub-25ms-resume model is the cleanest economic fit in the category, and Webflow's preview workload is proof it holds in production. The trade is a closed, managed-only platform from a seed-stage vendor with a thin community footprint, against open-source incumbents that are actively closing the persistence gap.

Recommended for: AI SaaS teams running per-user, mostly-idle agent workspaces where wake latency is user-facing; teams that want sandboxes plus colocated agent and MCP hosting from one usage-billed vendor.

Not recommended for: BYOC/self-hosting requirements, GPU workloads, security teams that need an auditable isolation stack, or anyone who wants the category's largest ecosystem.

Outlook: Watch whether the persistence differentiator survives E2B's and Daytona's snapshot features, whether a hypervisor-level architecture disclosure lands, and whether the next round confirms the traction story — the "billions of requests" headline deserves independent verification it has not yet received.

---

Research by Ry Walker Research • methodology