# Chutes

Breakthrough Serverless Compute for AI, at Scale.

Source: https://wheredoirunthis.com/products/chutes · JSON: https://wheredoirunthis.com/api/products/chutes · Read on 2026-10-14

## At a glance

- Type: Serverless
- Kind: Open core
- Isolation: Container
- Free tier: No
- Starting price: $3/mo
- GPU: Yes
- Regions: 1
- Coverage: 13 of 61 assessed capabilities supported
- Agent readiness: read 10 of 11 signals, drive 10 of 11 signals

## Links

- Homepage: https://chutes.ai/
- Docs: https://chutes.ai/docs
- Pricing: https://chutes.ai/pricing
- Repository: https://github.com/chutesai/chutes-api

## Pricing

- Model: Pay-as-you-go token inference and GPU-second custom deployments, plus optional monthly request bundles.
- Unit: per token (inference) + per GPU-second/hour (deployed chutes) + per-deploy fee
- Free tier: No

| Tier | Price | Includes | Notes |
| --- | --- | --- | --- |
| Pay-as-you-go | Usage-based | Token inference priced by model; private RTX Pro 6000 $1.80/hour plus one-time $5.40 deploy fee. |  [source](https://chutes.ai/pricing) |
| Base | $3/month | 300 bundled requests/day; GPU deployment not included. |  [source](https://chutes.ai/llms.txt) |
| Plus | $10/month | 2000 bundled requests/day; GPU deployment not included. |  [source](https://chutes.ai/llms.txt) |
| Pro | $20/month | 5000 bundled requests/day; GPU deployment not included. |  [source](https://chutes.ai/llms.txt) |
| Enterprise | Custom pricing | Volume discounts, custom rate limits, dedicated support. |  [source](https://chutes.ai/pricing) |

- Overages: PAYG per-model token rates beyond bundled quotas; Plus discount 6%, Pro 10%; private GPU deployments billed separately.
- Card required: No
- Source: https://chutes.ai/pricing

## Website design

Desktop public website · assessed 2026-10-14

Complete assessment.

Dark, spacious AI infrastructure site with oversized pale sans-serif headings, mint pill actions, a rainbow abstract landing visual, terminal-style pricing panels and dense documentation navigation.

- Styles: Technical illustration, Corporate
- Landing page: captured
- Pricing: captured
- Docs: captured

## Runtime

- **GPU & accelerators**: Yes — "Self-serve TEE GPU: RTX Pro 6000, 96 GB, $1.80/hour; private deployment fee $5.40. API estimates for other GPU classes are not guaranteed self-serve capacity." [source](https://chutes.ai/pricing) (confidence: high)
- **BYOC (Bring your own cloud)**: na — "Not a bring-your-own-cloud model — compute comes from the permissionless Bittensor miner network (you can instead BECOME a miner to contribute GPUs, the inverse of BYOC)." (source: n/a (decentralized network)) (confidence: high)
- **Secrets management**: pass — "Create a secret (e.g. private HF token). POST /secrets/; list/get/delete endpoints require authentication." [source](https://chutes.ai/docs/api-reference/secret) (confidence: high)
- **Storage options**: Model weights/images handled by the platform; no user-facing object store or persistent volume documented. Inference is stateless; deployed chutes are ephemeral. — "Serverless, pay-per-use; storage of durable user data is not a documented first-class offering." [source](https://chutes.ai/docs/core-concepts/chutes) (confidence: medium)
- **Supported workloads**: llm-inference, custom-ai-apps (FastAPI chutes), image/diffusion-models, speech/video-models, embeddings, background/batch jobs — "'Deploy, scale, and run any AI model or workload' — LLM, image, speech, video models + custom AI applications; vLLM/SGLang/Diffusion templates." [source](https://chutes.ai/llms.txt) (confidence: high)
- **Persistence**: Ephemeral by default — serverless GPU compute, 'pay only for GPU time actually used', autoscaling (incl. scale-to-zero). Chutes are stateless FastAPI apps; durable state is the user's responsibility (no documented first-class persistent volume on the public surface). — "'Serverless deployment to GPU clusters ... Automatic scaling based on demand ... Pay-per-use billing.' No persistent-volume primitive documented publicly." [source](https://chutes.ai/docs/core-concepts/chutes) (confidence: medium)
- **Data jurisdiction**: na — "Decentralized, globally-distributed anonymous miners — no user-facing region/jurisdiction selection or data-residency control (a residency trade-off of the decentralized model; TEE addresses confidentiality rather than residency)." (source: docs) (confidence: medium)
- **Resource limits & quotas**: Per-user quotas; optional monthly request bundles; subscribed usage caps documented in API. — "Base $3/month 300 requests/day; Plus $10/month 2,000/day; Pro $20/month 5,000/day. These bundles are not guaranteed GPU capacity." [source](https://chutes.ai/llms.txt) (confidence: high)
- **SDK languages**: Python — "First-party Python SDK (`pip install chutes`) for building/deploying chutes; the inference surface is OpenAI-compatible so any OpenAI SDK (any language) can call it." [source](https://chutes.ai/docs/getting-started/installation) (confidence: high)
- **Network & egress controls**: partial — "The platform ENFORCES network containment on running chutes ('strictly limit ... network access') as a security measure — but this is platform-imposed, not a USER-configurable egress allowlist/policy." [source](https://chutes.ai/docs/core-concepts/security-architecture) (confidence: low)
- **Snapshots & forks**: fail — "Serverless ephemeral functions — no documented running-instance snapshot/fork primitive." (source: docs) (confidence: medium)
- **Deployment methods**: cli, sdk, api, container-image — "`chutes deploy my_chute:chute` (CLI) + the Python SDK (Chute/Image/NodeSelector) + REST API (api.chutes.ai) + custom Docker image build (Image().from_base(...))." [source](https://chutes.ai/llms.txt) (confidence: high)
- **Isolation**: Container — "Chutes library is injected into every chute container. Hardware confidential-compute protection adds Intel TDX/NVIDIA CC; current catalogue is all confidential_compute=true and private deployment selects tee=True. Container describes the application packaging, not an operator-trusted host boundary." [source](https://chutes.ai/docs/core-concepts/security-architecture) (confidence: medium)

## Capabilities

### AI & agents

| Capability | Status | Evidence |
| --- | --- | --- |
| Agent skills | Supported | "Use the plugin when Claude needs Chutes-specific skills, scripts, and MCP tools." [source](https://chutes.ai/agents) |
| AI app builder | Not supported | "Not a prompt-to-app builder — you deploy code (a FastAPI chute) via the SDK/CLI." [source](https://chutes.ai/) |
| Autonomous QA agent | Not applicable | "No autonomous app-QA agent (raw compute/inference platform)." [source](https://chutes.ai/) |
| Built-in coding agent | Not supported | "Chutes' agent products (chutesai/codex, openclaw, squad-api) are SEPARATE products built ON the platform, not a first-party coding agent inside the deploy flow." [source](https://chutes.ai/llms.txt) |
| BYOK for LLMs | Not supported | "You use Chutes' own models (or deploy your own chute); there is no bring-your-own external-provider-key surface." [source](https://chutes.ai/llms.txt) |
| Credential-injection proxy | Not supported | "No documented credential-injection-proxy capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |
| LLM gateway | Not supported | "Chutes IS the inference provider serving its OWN network catalogue (with sophisticated intra-catalogue failover routing: default:latency/throughput) — NOT a multi-PROVIDER gateway/proxy for your external-provider calls." [source](https://chutes.ai/llms.txt) |
| MCP client | Not supported | "No MCP client." [source](https://chutes.ai/) |
| MCP-server hosting | Not supported | "A chute is an arbitrary FastAPI app (so an MCP server COULD be deployed as one), but hosting MCP servers is not an advertised feature." [source](https://chutes.ai/docs) |
| MCP server | Supported | "The toolkit ships a stdio MCP server for model discovery, quota, aliases, usage, discounts, chutes, and API key read paths. Write and deploy paths keep their [BETA] labels." [source](https://chutes.ai/agents/mcp) |
| Model inference API (first-party) | Supported | "Fresh public JSON catalogue contains model IDs, per-token price/pricing and confidential_compute=true." [source](https://llm.chutes.ai/v1/models) |
| Preinstalled coding agent | Not supported | "No documented preinstalled-coding-agent capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |

### Networking & delivery

| Capability | Status | Evidence |
| --- | --- | --- |
| Automatic HTTPS | Supported | "API + deployed chutes are served over HTTPS (e.g. https://myuser-my-llm.chutes.ai/v1) — automatic TLS endpoints." [source](https://chutes.ai/llms.txt) |
| Custom domains | Not supported | "Chutes get *.chutes.ai subdomains; no bring-your-own custom domain documented." [source](https://chutes.ai/docs) |
| Direct public IPv6 | Not applicable | "Serverless functions behind a managed HTTPS endpoint — no per-instance public IPv6." [source](https://chutes.ai/docs) |
| Edge KV store | Not applicable | "No edge KV store." [source](https://chutes.ai/) |
| Egress controls | Partially supported | "Platform ENFORCES network containment on running chutes ('strictly limit ... network access') — security policy, not a user-configurable egress allowlist." [source](https://chutes.ai/docs/core-concepts/security-architecture) |
| Email send / receive | Not applicable | "Not an email product." [source](https://chutes.ai/) |
| Public preview URLs | Not supported | "Deploy registers your chute and brings it online as private by default. A dedicated production endpoint is not an advertised temporary preview URL." [source](https://chutes.ai/pricing) |
| Private networking | Not applicable | "Serverless functions on a decentralized network — no user service-to-service private mesh/VPC." [source](https://chutes.ai/docs) |
| Built-in reverse proxy | Partially supported | "The validator routes external requests to your chute running on a miner (platform-managed routing) — not a user-configurable ingress/proxy." [source](https://chutes.ai/docs/core-concepts/security-architecture) |

### Organisation & security

| Capability | Status | Evidence |
| --- | --- | --- |
| Built-in auth | Partially supported | "'Sign in with Chutes' — a first-party OAuth 2.0 + OIDC IdP (create OAuth apps, client_id/secret, PKCE) for apps built on Chutes. An IdP for the platform; not a full end-user IAM for arbitrary workloads." [source](https://chutes.ai/docs/sign-in-with-chutes/overview) |
| BYOC | Not supported | "No run-in-your-own-cloud — compute is the permissionless miner network (you can instead provide GPUs as a miner)." [source](https://chutes.ai/docs) |
| Deploy into customers' VPCs | Not supported | "No documented customer-vpc-deploy capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |
| Secrets management | Supported | "Create a secret (e.g. private HF token)." [source](https://chutes.ai/docs/api-reference/secret) |
| Self-host / OSS core | Supported | "MIT open core: chutesai/chutes-api (backend platform API), chutesai/chutes (SDK), chutesai/chutes-miner (GPU provider). Self-hostable (caveat: no decentralized GPU supply of your own; some security components closed)." [source](https://github.com/chutesai/chutes-api) |
| Sharing | Partially supported | "Private access controls for every deployment; public model catalogue exists, but private-by-default custom deployment is not team collaboration." [source](https://chutes.ai/custom-chutes) |
| SSO / SAML | Not supported | "OAuth/OIDC sign-in exists, but no enterprise SAML SSO documented." [source](https://chutes.ai/docs) |
| Teams | Not supported | "Team-oriented marketing and private access controls, but no documented organization/team account-management interface on inspected sources." [source](https://chutes.ai/custom-chutes) |
| Workload identity | Not supported | "No documented workload-identity capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |

### Developer access

| Capability | Status | Evidence |
| --- | --- | --- |
| CLI | Supported | "Official Chutes CLI (`chutes deploy`, `chutes register`) — installed with the SDK." [source](https://chutes.ai/docs/cli/overview) |
| E2B-compatible API | Not supported | "No documented e2b-compatible-api capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |
| Filesystem API | Not applicable | "A chute defines its own FastAPI endpoints (cords); no remote filesystem-ops API like a sandbox product." [source](https://chutes.ai/docs) |
| Git deploy | Not supported | "Deploy is via CLI/SDK (`chutes deploy`), not git push-to-deploy." [source](https://chutes.ai/docs) |
| HTTP data API | Not applicable | "Not a data store — no HTTP data-query API." [source](https://chutes.ai/) |
| OpenAPI spec | Supported | "Full OpenAPI spec fetchable at api.chutes.ai/openapi.json (243KB) + Swagger UI." [source](https://api.chutes.ai/openapi.json) |
| Private container registry | Not supported | "Custom Docker images are built via the SDK Image() builder; no user-facing private container registry SKU documented." [source](https://chutes.ai/docs/core-concepts/chutes) |
| Public API | Supported | "Comprehensive documented REST API (api.chutes.ai) with Swagger UI — keys, chutes, billing, usage, models." [source](https://api.chutes.ai/docs) |
| SDK | Supported | "First-party Python SDK (`pip install chutes`) + e2ee transport plugin; OpenAI-compat means any OpenAI SDK works for inference." [source](https://chutes.ai/docs/getting-started/installation) |
| SSH access | Not applicable | "Serverless functions — no SSH into the running workload." [source](https://chutes.ai/docs) |

### Compute & storage

| Capability | Status | Evidence |
| --- | --- | --- |
| Cloud-init / ignition config | Not applicable | "Serverless — image customization is via the SDK Image() builder, not VM cloud-init/ignition." [source](https://chutes.ai/docs/core-concepts/chutes) |
| Desktop / computer-use sandbox | Not applicable | "No virtual desktop / computer-use." [source](https://chutes.ai/) |
| GPU compute | Supported | "RTX Pro 6000; 96 GB; $1.80 / hour; One-time deployment fee $5.40." [source](https://chutes.ai/pricing) |
| Live-state (memory) snapshots | Not supported | "No documented memory-snapshot capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |
| Multi-region | Partially supported | "Globally distributed across permissionless Bittensor miners (inherently multi-region), BUT no user-facing region selection or data-residency control." [source](https://chutes.ai/docs/core-concepts/security-architecture) |
| Object storage | Not supported | "No user-facing S3-compatible object store offered to chutes." [source](https://chutes.ai/docs) |
| Persistent disk | Not supported | "Jobs mention persistent storage for results, but examples use local /tmp files; no managed persistent disk or user volume contract established." [source](https://chutes.ai/docs/core-concepts/jobs) |
| Sandbox templates / custom images | Supported | "Custom Docker image building (Image().from_base()) + first-party vLLM / SGLang / Diffusion templates to launch model chutes from." [source](https://chutes.ai/docs/templates/vllm) |
| Scale to zero | Supported | "Idle instances shut down automatically, so no GPU-seconds are wasted." [source](https://chutes.ai/pricing) |
| Selectable isolation runtime | Not supported | "No documented selectable-isolation-runtime capability on the inspected API, SDK and deployment surface; custom GPU workloads are not a sandbox compatibility or workload identity product." [source](https://chutes.ai/docs) |
| Snapshots / forking | Not supported | "Serverless functions — no running-instance snapshot/fork primitive." [source](https://chutes.ai/docs/core-concepts/chutes) |

### Data

| Capability | Status | Evidence |
| --- | --- | --- |
| Database branching | Not applicable | "Not a data layer." [source](https://chutes.ai/) |
| Ephemeral instant DB (no-signup) | Not applicable | "Data-layer primitive — not applicable." [source](https://chutes.ai/) |
| First-party database | Not supported | "No managed database offered to user workloads (the platform uses Postgres internally)." [source](https://chutes.ai/docs) |
| Vector search | Not applicable | "No first-party vector/embedding store (you could deploy an embedding chute, but it's not a platform feature)." [source](https://chutes.ai/) |

### Platform & operations

| Capability | Status | Evidence |
| --- | --- | --- |
| Managed CI/CD runners | Not applicable | "Not a CI-runner platform." [source](https://chutes.ai/) |
| Built-in observability | Partially supported | "Instance log streaming (GET /instances/{id}/logs) + invocation usage/stats APIs (per-model TPS/TTFT) — metrics+logs, no full tracing/APM stack." [source](https://chutes.ai/llms.txt) |
| Preview environments | Not applicable | "No per-PR ephemeral environments (no git-based PR flow)." [source](https://chutes.ai/docs) |
| Scheduled tasks | Not supported | "No cron/scheduled-job primitive documented." [source](https://chutes.ai/docs) |
| Project template marketplace | Partially supported | "First-party deploy templates (vLLM / SGLang / Diffusion) + a public catalogue of community chutes to use — deploy-starter templates, not a full clone-and-customize project marketplace." [source](https://chutes.ai/docs/templates/vllm) |
| Durable workflows | Partially supported | "Jobs: Run for extended periods (hours or days); Store results in persistent storage; Retry on failure automatically. No checkpoint/replay workflow engine established." [source](https://chutes.ai/docs/core-concepts/jobs) |

## Agent readiness

### Read: 10 of 11 signals present

10/11 signals (8 full, 2 partial)

- Llms txt: pass — "Fresh /llms.txt links models, pricing, agent setup, full docs, auth and legal pages."
- Llms full txt: pass — "HTTP 200 comprehensive llms-full.txt export; 1247193 bytes, retained via pointer."
- Robots AI policy: pass — "Fresh robots.txt explicitly allows GPTBot, OAI-SearchBot, ClaudeBot and other crawlers, excluding private routes."
- JSON LD: partial — "Homepage contains one WebSite JSON-LD with Organization publisher, not a rich Product/SoftwareApplication."
- Sitemap: pass — "Fresh sitemap.xml HTTP 200; 209 URL locations."
- Meta quality: pass — "Descriptive <title> + meta description + 10 OG tags."
- Markdown negotiation: fail — "Docs requested with Accept: text/markdown still returns text/html; no clean negotiation established."
- Content efficiency: partial — "Readable homepage text / HTML bytes = 2.10%, using extracted non-script/style HTML text."
- Docs discoverable: pass — "First-class docs (chutes.ai/docs) + Swagger (api.chutes.ai/docs) + a knowledge base, all linked."
- Pricing machine readable: pass — "GET /pricing returns structured per-GPU JSON; GET /v1/models returns per-token pricing (USD+TAO) — genuinely machine-readable, better than most."
- Anti bot: pass — "Anonymous public landing, docs, pricing, robots, exports and spec fetched without challenge."

### Drive: 10 of 11 signals present

10/11 signals (8 full, 2 partial)

- Public API: pass — "Documented REST API (api.chutes.ai) + Swagger UI — full control plane."
- Openapi spec: pass — "Fresh HTTP 200 API OpenAPI JSON, 293083 bytes."
- MCP server: pass — "Official /agents/mcp documents uv tool install chutes-mcp-server --from local toolkit directory, stdio client configuration and read tools; writes/deploy are BETA."
- SDK packages: partial — "One official deployment SDK language: Python pip install chutes. Generic OpenAI clients are compatible, not multiple official Chutes SDKs."
- Auth docs: pass — "Authentication API reference and llms.txt document Bearer key authentication and OAuth identity provider."
- Agent self serve creds: pass — "Fresh OpenAPI lists POST /users/register and /api_keys/; toolkit documents autonomous agent registration. Documentation only, no registration performed."
- Webhooks: fail — "Inspected fresh OpenAPI/docs show no signed public webhook integration."
- CLI tool: pass — "Official `chutes` CLI (deploy/register)."
- Error model: pass — "OpenAPI defines HTTPValidationError and documented 422 JSON validation responses."
- Rate limit docs: partial — "Quota endpoints and request bundles documented, but no comprehensive response-header contract established."
- Agent config: pass — "Official agent hub publishes skills, scripts and MCP installation with multiple agent-specific guides."

## Coding harnesses

- Claude Code (Anthropic) — "Claude Code and Cowork: setup, plus Claude recipes" [source](https://chutes.ai/llms.txt)
- Codex CLI (OpenAI) — "Codex: guide-only setup, plus Codex recipes" [source](https://chutes.ai/llms.txt)
- Cursor (Anysphere) — "Cursor, Cline, Aider: editor agents" [source](https://chutes.ai/llms.txt)
- Cline (Cline) — "Cursor, Cline, Aider: editor agents" [source](https://chutes.ai/llms.txt)
- Aider (Aider) — "Cursor, Cline, Aider: editor agents" [source](https://chutes.ai/llms.txt)
- Kilo Code (Kilo) — "Confirmed live integrations: LiteLLM, Vercel AI SDK, LangChain, Cline, Kilo Code, Roo Code, Cursor, Aider, MCP." [source](https://chutes.ai/llms.txt)
- Roo code — "Confirmed live integrations: LiteLLM, Vercel AI SDK, LangChain, Cline, Kilo Code, Roo Code, Cursor, Aider, MCP." [source](https://chutes.ai/llms.txt)
- Hermes Agent (Nous Research) — "Hermes: setup, plus Hermes recipes" [source](https://chutes.ai/llms.txt)
- OpenClaw (Peter Steinberger) — "OpenClaw: provider setup [BETA], doc-derived. Compatibility documented, not exercised in this scan." [source](https://chutes.ai/llms.txt)

## Compliance & regions

- GDPR: self-attested (scope: hosted) — "This agreement ensures compliance with applicable data protection laws, including the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA)." [source](https://chutes.ai/dpa)
- Regions: Decentralized global miner network (Bittensor) (worldwide, permissionless miners)
- Regions note: Compute is supplied by a globally-distributed, permissionless miner network — inherently multi-region but with NO user-facing region selection or data-residency control (a trade-off of the decentralized model; live node/utilization data at chutes.ai/app/research/nodes). The hosted control plane (api.chutes.ai / llm.chutes.ai) is centralized; the GPU execution is decentralized.

## Open source

- Repository: https://github.com/chutesai/chutes-api
- License: MIT
- Stars: 28
- Forks: 18
- Self-host methods: docker, compose, kubernetes, helm, ansible
- Self-host dependencies: postgres, kubernetes, gpu

## Company & presence

- Headquarters: Distributed / decentralized (no central HQ)
- Founded: 2024 (under Chutes Global Corp; a Bittensor-subnet-native platform)
- Team size: 8 named people on current team page; not a total employee headcount.
- Privacy policy: https://chutes.ai/privacy
- Terms of service: https://chutes.ai/terms
- Website: https://chutes.ai/
- Docs: https://chutes.ai/docs
- GitHub (chutesai org): https://github.com/chutesai
- News / Blog: https://chutes.ai/news
- Knowledge Base: https://chutesai.zohodesk.com/portal/en/kb/chutes-ai
- Chutes Chat: https://chutes.ai/chat
- Chutes Search: https://search.chutes.ai/

## Nearby in the atlas

- [Beam](/products/beam.md)
- [Deno Deploy](/products/deno.md)
- [Koyeb](/products/koyeb.md)
- [Kyushu](/products/kyushu.md)
- [Modal](/products/modal.md)
- [NVCF (NVIDIA Cloud Functions)](/products/nvcf.md)

## Method

Facts come from the product's own public pages. Each claim carries its evidence: a quote and the page it came from. We count capabilities rather than grade products, and "Not assessed" means we have not checked, never that the product lacks it.