Runta
Secure runtimes for AI agents that reduce token usage, keep credentials out of prompts, and govern network access.
At a Glance
Engagement
Available On
Alternatives
Listed Nov 2026
About Runta
Runta is an execution layer for AI agents that runs them in managed runtimes with built-in controls for token spend, credentials, and network egress. Users create runtimes from a dashboard, CLI, or Python and TypeScript SDKs, and can start coding agents such as Codex CLI, Claude Code, or OpenClaw inside them. The company says its team built boundary infrastructure at Cloudflare and Kong.
What It Is
Runta provides sandboxed runtimes where agents execute commands, with controls applied at the layer the agent runs on. Runtimes are created with a name and compute resources (vCPUs and memory), and commands run inside them through the CLI or SDKs. Agent presets install and configure a chosen coding agent automatically.
Spend, Control, and Scale
The homepage groups capabilities into three areas:
- Spend: prompt compression shrinks oversized tool output such as JSON arrays, logs, search results, and git diffs while keeping errors, important items, and changed lines. Token X-ray is documented alongside it.
- Control: secrets are stored once and injected into matching requests, such as OpenAI API calls, so the real key is not exposed inside the runtime. Governed network egress allows only approved access.
- Scale: auto suspend and resume, and memory auto-scaling.
Setup Path
Create an API key under Settings → Runta API Keys in the dashboard and export it as RUNTA_TOKEN. Install the CLI (npm on Linux, Homebrew on macOS Apple silicon) or the Python or TypeScript SDK. Setup currently targets macOS and Linux; Windows is not supported yet. The docs also offer installable skills and a prompt for setting up a Pi coding agent, plus a Cloud Agents quickstart for managed agents with persistent computers.
Community Discussions
Be the first to start a conversation about Runta
Share your experience with Runta, ask questions, or help others learn from your insights.
Pricing
Usage-based
Prepaid, usage-based billing with no subscription and no monthly minimum; minimum manual top-up is $5.
- vCPU $0.0504 per vCPU-hour
- Memory $0.0162 per GiB-hour
- Disk storage $0.000108 per GiB-hour
- Running: vCPU, memory, and disk storage billed
- Paused: memory and disk storage; no vCPU charges
- Suspended or shut down: disk storage only
- Deleted: nothing
- Minimum manual top-up $5
- No subscription, no monthly minimum
Capabilities
Key Features
- Prompt compression for structured tool output
- Token X-ray
- Credential protection with secret injection
- Governed network egress
- Auto suspend and resume
- Memory auto-scaling
- Agent presets for Codex CLI, Claude Code, OpenClaw, and Pi
- CLI, Python SDK, and TypeScript SDK
- Dashboard runtime creation
- Cloud Agents with persistent computer and streaming API
