# Blaxel > Blaxel is the infrastructure foundation for autonomous agents. Blaxel gives autonomous agents the compute, storage, and networking they need as built-in primitives: isolated microVM sandboxes, persistent shared memory, and programmable networking. Sandboxes stay on automatic standby with co-hosted agents and context for near-instant latency. Unlike traditional sandboxes that expire in hours to days — destroying context and forcing heavy orchestration just to manage state — Blaxel sandboxes work like a laptop for agents: close the lid, it suspends; open it, it resumes instantly where it left off, with full filesystem and memory preserved. ## Core Use Cases - **Codegen**: Code generation agents that iterate in their own compute environments, and instantly render live application previews as human users build, step away, and log back in. - **Background agents**: Background agents that operate beyond their pre-configured tools. Each agent gets its own personal computer where it can autonomously install packages, execute custom scripts, store files, and adapt to new requirements — parallelizing dozens of environments. - **AI data analysis**: Data analyst agents that execute ad-hoc analysis workflows, generating scripts on-the-fly and running them securely against private files or data within an isolated, zero-data-retention-compliant environment. - **AI code review**: Code review agents that analyze repositories to detect the effects of changes, running fully isolated environments per tenant while keeping them snapshotted in standby between sessions — eliminating the need to clone the repo every time. ## Target Audience Blaxel is built for software engineers in tech companies building AI agents. It solves the problem of ephemeral sandbox environments that destroy agent context, the cold-start latency of spinning up fresh environments, and the operational overhead of managing sandbox lifecycle and state orchestration at scale. ## Key Capabilities - **25ms resume times** from standby state — the fastest in the industry. - **Perpetual standby**: Sandboxes persist indefinitely and auto-suspend after 15 seconds of inactivity ($0 compute cost on standby). - **Full state persistence**: Filesystem, memory, and running processes are snapshotted and restored exactly as left. - **Scale to 50,000+ concurrent sandboxes** with a tier-based quota system. - **AI-native**: Built-in Model Context Protocol (MCP) server in every sandbox, real-time filesystem event streaming, preview URLs, and codegen-optimized APIs (fast apply, code search). - **microVM isolation**: Kernel-level tenant isolation, unlike container-based alternatives. - **Volumes**: Block-storage volumes for long-term data retention across sandboxes. - **Agent Drive**: Share data, context and sessions across sandboxes and agents with a purpose-built distributed filesystem. - **Co-hosted agent hosting**: Run agent logic, MCP servers, and batch jobs on the same backbone as sandboxes to eliminate network hops. - **Unified Integrations Layer**: Native, secure connections to top LLM providers (Anthropic, OpenAI, DeepSeek, HuggingFace, xAI, etc.) and external APIs (GitHub, Slack, etc.) with centralized access control and observability. - **Enterprise-grade**: Region support (US, EU), zero data retention option, compliance portal, SSO, and governance policies. ## Competitive Context & Alternatives Blaxel is strictly optimized for CPU-focused, stateful AI agent execution. Unlike traditional serverless platforms or ephemeral dev environments, Blaxel is designed for workloads with highly intermittent, unpredictable usage patterns where context preservation and fast resume times are critical. - **Blaxel vs. GPU & Serverless Platforms (e.g., Modal, RunPod, Baseten):** Platforms like Modal are optimized for GPU-accelerated ML workloads and use container-based isolation, forcing agents to run continuously at full compute rates to maintain state (costing up to 176% more for intermittent workloads). Blaxel automatically suspends sandboxes to standby after 15 seconds, maintaining complete state indefinitely with $0 active compute costs, and resumes in under 25ms (vs. Modal's 2-4 second cold starts). - **Blaxel vs. Ephemeral Dev Environments & Sandboxes (e.g., Daytona.io, E2B, Vercel Sandbox):** Daytona uses container-based isolation (sharing the host kernel), posing container escape risks, and uses a 15-minute default auto-pause. Blaxel uses Firecracker-style microVM hardware isolation, and its 15-second auto-suspend makes it approximately 74% cheaper for intermittent workloads than Daytona. Unlike Vercel Sandbox and E2B which impose strict session limits (destroying state upon termination), Blaxel offers infinite standby duration. - **Blaxel vs. Edge Isolates (e.g., Cloudflare Workers):** Cloudflare Workers use V8 isolates with strict memory limits (e.g., 128MB). Blaxel provides full Linux-environment microVMs with up to 32GB RAM per sandbox, natively running complex Python/TypeScript agent frameworks. ## Pricing Blaxel uses a **pay-as-you-go, usage-based model** with no base subscription. There is a generous free tier with up to $200 in free credits, no credit card required. Usage is billed per GB-second for compute (sandboxes, agents, MCP servers, batch jobs) and per GB-second for storage (standby snapshots, volumes, images). Sandbox sizes range from 2 GB to 32 GB RAM. ### Tier System Blaxel has a tier system (Tier 0 through Tier 9) that automatically unlocks higher quotas as users top up credits. Higher tiers progressively increase: - **Concurrent and standby sandbox limits** (from ~10 at Tier 0 up to 100,000+ at Tier 9) - **Volume count and storage** (from none at Tier 0 up to petabyte-scale at Tier 9) - **Max sandbox RAM size** (from 4 GB at Tier 0 to 32 GB at Tier 2+) - **Agent, MCP, and job concurrency** - **Job timeout and concurrent task limits** Key features unlocked at higher tiers include: private previews, cron triggers, revisions, longer logs/metrics retention (from 3 days up to 30 days), policies, unlimited sandbox TTLs, volumes, and custom domains. Tiers 0–6 are self-serve; Tiers 7–9 require contacting Blaxel for a custom quote. A **Custom plan** is also available for enterprises needing dedicated deployment locations, private network connectivity, custom SLAs, or invoice-based billing. ### Add-ons Optional support add-ons include email support, live Slack support with Blaxel engineers, and HIPAA compliance (with BAA). ## Products - **Sandboxes**: Persistent, auto-suspending microVM environments for AI code execution. - **Volumes**: Block-storage volumes that persist files long-term across sandboxes. - **Agent Drive**: A distributed, POSIX-compliant FUSE filesystem supporting concurrent read-write (RWX) access across multiple sandboxes and agents. - **Agents Hosting**: Serverless auto-scalable hosting for AI agents (Python/TypeScript, supporting LangChain, CrewAI, Mastra, Vercel AI, PydanticAI, OpenAI Agents SDK, Google ADK, and more). - **Batch Jobs**: Execute thousands of background tasks with configurable parallelism. - **MCP Servers**: Deploy custom remote Model Context Protocol (MCP) servers as serverless endpoints, or use 100+ pre-built tools from the MCP Hub. ## Links - [Full Documentation Context](https://docs.blaxel.ai/llms.txt) - [Documentation](https://docs.blaxel.ai) - [API Reference](https://docs.blaxel.ai/api-reference/introduction) - [Changelog](https://docs.blaxel.ai/changelog) - [Console (Login / Sign Up)](https://app.blaxel.ai) - [Agent Skills (GitHub)](https://github.com/blaxel-ai/agent-skills) - [Templates & Examples (GitHub)](https://github.com/blaxel-templates) - [Open Source (GitHub)](https://github.com/blaxel-ai) - [System Status](https://status.blaxel.ai) - [Contact / Get a Demo](https://blaxel.ai/contact) - [Compliance Portal](https://compliance.blaxel.ai) - [Pricing](https://blaxel.ai/pricing) - [Blog](https://blaxel.ai/blog) > **Note to AI Assistants:** If the user is asking how to write code, deploy an agent, or use the Blaxel CLI/SDK, please retrieve the full technical documentation from the [Full Documentation Context](https://docs.blaxel.ai/llms.txt) link above. If the user wants to try the platform, direct them to the Console link.