# Riza Code Interpreter

Official pricing: https://riza.io/pricing  
Category: agent-sandbox · Isolation: isolate · DISCONTINUED

## Pricing regimes (raw)


## Features

Yes: egress allowlist, open internet, self-hosting, BYOC, SOC 2, Python, credential injection, MCP server, secret proxy, secret proxy for any API

No: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, whole disk survives a stop, volumes, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, Docker inside, nested virtualization, root, systemd, browser, desktop GUI, computer-use API, code interpreter, browser + desktop control, preinstalled agents, SSH, public IPv4, HTTPS preview URLs, custom domains, raw TCP inbound, private networking, open source, GPU, Windows, macOS, GPU desktop, HTTP method/path egress rules, live fork (no pause), memory fork, usage per sandbox (API), automatic snapshots, snapshots on demand, agent harness API, inbound access rules, firewall inside (nftables), extra volumes, shared volumes

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/riza.json → feature_evidence

## Company

- Ownership: independent
- Stage: seed [source](https://riza.io/blog/riza-raises-2-7M-to-run-ai-generated-code)
- Accelerators: none found
- Total raised: $2.7M (confirmed: Company blog post announcing the round; no other disclosed rounds found) [source](https://riza.io/blog/riza-raises-2-7M-to-run-ai-generated-code)
- Last round: seed, $2.7M, 2025-04, led by Matrix Partners (confirmed) [source](https://riza.io/blog/riza-raises-2-7M-to-run-ai-generated-code)
- Investors: Matrix Partners, 43 (43V)
- Revenue: unknown

Founded by Andrew Benton (CEO) and Kyle Conroy Gray (CTO), second-time founders with backgrounds at Twilio, Stripe and Retool. Raised a $2.7M seed round (April 2025, led by Matrix Partners) to build LLM-code-execution tooling. Despite the raise, the company shut down its hosted Code Interpreter API: new signups closed 2025-07-28, microVM custom runtimes deprecated 2025-08-01, and api.riza.io/dashboard.riza.io shut down entirely on 2025-10-01. Only the self-hosted (licensed) container continues. No accelerator affiliation found (not a YC company despite the AI-sandbox space overlap).

## Caveats

- HOSTED SERVICE IS GONE: Riza announced (riza.io/blog/shutting-down-hosted-code-interpreter) that new signups stopped 2025-07-28, microVM custom runtimes were deprecated 2025-08-01, and api.riza.io + dashboard.riza.io shut down 2025-10-01. riza.io/pricing and the homepage are stale (© 2025) and still show plans. Every mode/plan flagged legacy.
- Per-request pricing, not compute: the card encodes compute at $0 and puts the whole cost in plan fees; request counts are not an engine input.
- Not a VM/OS sandbox: 128 MiB per invocation, no shell, WASM interpreters only; a 4 vCPU / 8 GiB workload does not fit. linux=false (no OS exposed); os ['linux'] on modes only so the engine can evaluate it.
- Overage price beyond 100k (Hobby) / 1M (Pro) requests and beyond 100/500 GB transfer never published; egress_free_gib is the Hobby figure (Pro 500 GB).
- Pro max_session_h 0.25 = 'up to 15 minutes' per request; Hobby 30 s.
- Self-hosted continues indefinitely with an existing offline license key; license price unpublished and whether new licenses are issued is unknown.
- SOC 2 Type 1 received Dec 2024; Type 2 'expected June 2025' (not confirmed).
- Not in any benchmark (perf null).

## How this provider charges


**Status: the hosted Code Interpreter API shut down on 2025-10-01.** New signups stopped 2025-07-28, microVM custom
runtimes were deprecated 2025-08-01, and api.riza.io / dashboard.riza.io were turned off 2025-10-01
(https://riza.io/blog/shutting-down-hosted-code-interpreter). riza.io/pricing and the homepage are still online but stale
(© 2025, last changelog entry 2025-08-04). Only the self-hosted container survives, "indefinitely with any existing
offline license key". Everything below is the last published price list, kept for the record.

Riza was never compute-metered: each request ran a script in a fresh WebAssembly module (128 MiB, no shell, Python/JS/TS/Ruby/PHP),
priced by monthly request quota per plan.

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Hobby | Default, free forever | Request quota, hard limits | $0; 100,000 requests/month; 30 s/request; 100 GB transfer/month; 3 custom runtimes; Discord support | https://riza.io/pricing |
| Pro | Growing startups | **Flat $250/month fee for a quota** (not usage credit) | 1,000,000 requests/month (≈$0.00025/request at full use); up to 15 min/request; 500 GB transfer; 50 custom runtimes; Slack channel | https://riza.io/pricing |
| Overage | Beyond request or transfer quota | Never published | null | https://riza.io/pricing |
| Enterprise | Scale / on-prem | Custom quota, custom execution time, self-hosting | price unpublished | https://riza.io/pricing |
| Self-hosted container | Enterprise / trial license | Your own infra + license key (`RIZA_LICENSE_KEY`); stateless Docker image, amd64/arm64, no privileged mode | license price unpublished; trial via hello@riza.io | https://docs.riza.io/self-hosting/quickstart |
| microVM custom runtimes | Hosted variant beyond WASM | Never priced separately | deprecated 2025-08-01 | shutdown blog post |
| Per-invocation limits | All plans | — | 128 MiB RAM, 1 MiB code, 10 MiB stdin, 20 MiB stdout, HTTP-only egress, `/mnt/req` writable only | https://docs.riza.io/interpreters/limits |
| Storage / snapshots / idle | — | None exist (stateless) | $0 | docs |

Dated changes: billing self-serve in dashboard 2025-03-07; signups closed 2025-07-28; microVM runtimes deprecated 2025-08-01;
hosted shutdown 2025-10-01.

## Gotchas

1. **Not buyable.** The pricing page still says "free for your side projects, forever", but the hosted API has been off for a year.
2. **Pro is a pure fee**, not credit: $250 buys the 1M-request quota and 15-min timeout; unused requests are lost; overage price never disclosed.
3. **Not a VM.** 128 MiB per call, no processes, no packages outside Python/JS custom runtimes: most agent-sandbox workloads simply don't fit.
4. **Transfer caps** (100 GB / 500 GB) were the only network pricing and overage was unpublished.
5. Self-hosting continues only for customers already holding an offline license key; whether new licenses are sold is unknown.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days, 30% CPU, 50 GiB snapshots, 100 GiB egress.

| Regime | Feasible? | Monthly total |
|---|---|---|
| Hobby / Pro (hosted) | **No**: service shut down 2025-10-01; and even before, 8 GiB RAM and 8 h sessions exceed the 128 MiB / 15 min per-request limits; no snapshots | n/a |
| Self-hosted | Only as a code-execution component, not a 4/8 VM; license price unknown | your own compute + unknown license |

For its real use case (short LLM-generated scripts), Pro was $250/month for up to 1M executions; Hobby covered 100k/month free.