Build agents.Finish anything.
CleverThis is the agent command center for everyone shipping with LLMs — and everyone using agents to get real work done. Develop, version, deploy, and orchestrate any agent locally or in the cloud, with hierarchical agents and hierarchical actors purpose-built for the long-running, multi-step jobs that flatten single-shot agents. Think of your favorite AI coding tool — minus the “coding-only” ceiling. Self-hostable. MIT-licensed.
Free tier: 1,000 requests / month. No card required.curl -s https://dev.cleverthis.com/v1/chat/completions \
-H "Authorization: Bearer $CLEVERTHIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-4o",
"messages": [{"role": "user", "content": "Hello!"}]
}' \
| jq '{content: .choices[0].message.content, provider: .provider, request_id: .cleverthis.request_id}'Built by the team behind
From zero to first call in three minutes
No new SDK, no proprietary protocol, no migration script. Sign up, set one environment variable, and your existing OpenAI client (or plain cURL) is now talking to every model provider through CleverThis.
Sign up, grab a key
Free tier: 1,000 requests / month. No card required. Your API key is created automatically on signup and shown once.
List available models
Every provider CleverThis supports is exposed through the same OpenAI-compatible model list. One call shows you what is available and what pricing applies.
curl -s https://dev.cleverthis.com/v1/models \
-H "Authorization: Bearer $CLEVERTHIS_API_KEY" \
| jq '.data[] | {id: .id, provider: .owned_by, available: .cleverthis.available}'Make the first call
The gateway routes to the provider you specify, meters the request through OpenMeter OSS, and returns provenance — provider, cost, and request ID — in the same response body.
curl -s https://dev.cleverthis.com/v1/chat/completions \
-H "Authorization: Bearer $CLEVERTHIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-4o",
"messages": [
{"role": "user", "content": "What is 2+2?"}
]
}' | jq '{content: .choices[0].message.content, provider: .provider, cost: .usage.cost, request_id: .cleverthis.request_id}'Features
Each one is a real, working capability — click a tab to see the snippet. Every behaviour shown here is documented in the public specification.
OpenAI-compatible API
Every supported provider — OpenAI, Anthropic, Google, Bedrock, Azure, and any BYO endpoint — is accessed through the same POST /v1/chat/completions shape. Streaming, tool calls, JSON mode, and structured outputs all work unchanged.
curl -s https://dev.cleverthis.com/v1/chat/completions \
-H "Authorization: Bearer $CLEVERTHIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-3-5-sonnet",
"messages": [
{"role": "user", "content": "Write a haiku about recursion."}
],
"stream": true,
"temperature": 0.7
}'OpenAI-compatible API
Every supported provider — OpenAI, Anthropic, Google, Bedrock, Azure, and any BYO endpoint — is accessed through the same POST /v1/chat/completions shape. Streaming, tool calls, JSON mode, and structured outputs all work unchanged.
curl -s https://dev.cleverthis.com/v1/chat/completions \
-H "Authorization: Bearer $CLEVERTHIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-3-5-sonnet",
"messages": [
{"role": "user", "content": "Write a haiku about recursion."}
],
"stream": true,
"temperature": 0.7
}'Why CleverThis
Eight things that break the moment you try to run agents seriously. Eight things CleverThis solves at the foundation — every one of them shipped today and documented in the public design record.
Hierarchical agents that actually finish
Single-shot agents bloat their context, retry forever, and stall before the job is done. Hierarchical agents and hierarchical actors decompose hour-long, multi-step work into isolated subtasks — each with its own context window — so the top-level agent stays sharp and the work actually ships.
Develop with CleverAgents. Ship to CleverThis.
CleverAgents is the open-source CLI that mirrors the CleverThis runtime — same actor format, same hierarchical execution, same provider surface. Prototype on your laptop with local models, then push the same actor bundles up to CleverThis when you need scale, observability, and managed infrastructure. The artifacts move unchanged.
Every model. One token. No SDK rewrites.
Swap a single environment variable and your existing OpenAI SDK speaks to OpenAI, Anthropic, Google, Bedrock, Azure, Ollama, vLLM, or anything OpenAI-compatible you point at. Adding a new provider stops being a refactor and starts being a config change.
Versioned, content-addressed agents
Every agent has an immutable SHA-based identity computed from its canonical form. Identical definitions produce identical IDs. Published agents cannot be retroactively altered — so “which version produced that output?” has an answer, every single time.
A real package registry for everything you build
Actors, skills, MCP servers, prompt templates, and graph definitions are first-class versioned packages — not files scattered across notebooks, Slack threads, and someone’s laptop. Share, compose, and depend on agent components the same way you depend on code libraries.
Failover before your users notice
Define an ordered list of providers per request. When the primary 429s, 500s, or rate-limits mid-stream, CleverThis transparently retries the next candidate inside the same response stream. Your users see one answer. You get the telemetry on what actually happened.
Hard spend limits, enforced in real time
A local transactional ledger reserves funds against your wallet before every upstream call. Insufficient balance? The request is rejected with HTTP 402 — before a single provider token is spent. Runaway agents burn nothing they aren’t allowed to.
Use any A2A client. Or our REST. Or our webapp.
Every operation the CleverThis webapp performs is a public A2A JSON-RPC or OpenAI-compatible REST call. Drive the platform from any A2A-compliant client, from our REST API, from the CleverAgents CLI, or from the webapp — you always get the same surface, the same data, and the same permissions. No private endpoints. No lock-in to our UI.
Open standards. Open-source spirit.
CleverThis is built and operated by an open-source shop. We ship with open protocols, publish what we can publish, and give developers honest tools that don’t hold their data or workflows hostage. Security and transparency are first-class concerns: every request produces an auditable trace, every public behaviour is documented in the spec, and every architecture change ships as a public ADR. You can read all of it before you trust any of it.
Built on standards we did not invent
A2A (Agent-to-Agent Protocol)
JSON-RPC control planeExtensible method namespace, Agent Card discovery. Any A2A-compliant client speaks to CleverThis without a custom integration.
CloudEvents
Usage event envelopeEvery metering substrate accepts CloudEvents natively. Your usage data lands wherever you can sink CloudEvents — no proprietary export.
JSON-RPC 2.0
RPC envelope over HTTPBattle-tested transport for the A2A control plane. No custom framing, no bespoke client SDK to learn.
LangChain
Universal LLM interfaceOne adapter family covers ~50 providers with uniform streaming, tool-calling, and token-usage extraction.
LangGraph
Bounded graph executorStateGraph, conditional edges, subgraphs, native streaming. The same runtime substrate the CleverAgents Actors library uses.
OAuth 2.0
Federated sign-inSign in with Google or GitHub. No password-management surface to attack when you don’t want one.
OpenAI API
Data-plane schemaEvery OpenAI client SDK works as-is — change one base URL and you’re routed through CleverThis.
OpenMeter OSS
Metering and entitlementsApache-2.0 metering layer, self-hostable. No cloud-only feature lock-in for billing aggregation.
OpenTelemetry
Distributed tracing and metricsVendor-neutral observability. Ship traces and metrics to any OTel-compatible backend you already run.
RFC 8785 (JSON Canonical Form)
Content-addressed actor identityDeterministic canonical bytes for every actor definition, so the same agent produces the same SHA-1 ID anywhere.
Stripe
Payments and checkoutIndustry-standard payment processing with webhook idempotency. No bespoke credit-card handling code on our side or yours.
CleverAgents — the MIT-licensed CLI counterpart
CleverThis the hosted platform is a managed service. But the CleverAgents CLI is open source and mirrors much of the CleverThis runtime — same actor format, same hierarchical execution, same provider surface. Use it to develop, run, and self-host agents locally with no account, no telemetry, no strings. Agent definitions move between CleverAgents and CleverThis unchanged.
What teams are saying
A glimpse of how teams use CleverThis in production. Have a quote you'd like to share? Get in touch.
“We swapped three provider integrations for a single CleverThis endpoint in an afternoon. The fallback logic alone has saved us from two outages this quarter — our customers didn't notice either of them.”
“The hosted Actor endpoints are the killer feature. We compile our prompts and chains down to IR, ship a stable URL, and our product team doesn't have to redeploy the app every time a prompt changes.”
“Explainable provenance per request was what closed the deal for us. Auditors get exactly what they need without us building a separate logging pipeline. The local self-host story is the icing on the cake.”
“Pay-as-you-go with no per-seat surprises, and a CLI that doesn't fight us in CI. We've been running CleverThis behind every internal LLM workload for six months. Boringly reliable.”
Pricing
Start free. Pay only for what you use. Enterprise plans for teams that need SSO, SLAs, or dedicated infrastructure.
Numbers below are preliminary — talk to us before you commit.
Free
$0
foreverPro
$0 base
pay-as-you-goEnterprise
Custom
contact usFrequently asked questions
- 302.AI
- Abacus
- abliteration.ai
- AIHubMix
- Alibaba
- Alibaba (China)
- Alibaba Coding Plan
- Alibaba Coding Plan (China)
- Amazon Bedrock
- Ambient
- Anthropic
- Atomic Chat
- Auriko
- Azure
- Azure Cognitive Services
- Bailing
- Baseten
- Berget.AI
- Cerebras
- Chutes
- Clarifai
- Claudinio
- Cloudflare AI Gateway
- Cloudflare Workers AI
- CloudFerro Sherlock
- Cohere
- Cortecs
- Databricks
- Deep Infra
- DeepSeek
- DigitalOcean
- DInference
- D.Run (China)
- evroc
- FastRouter
- Fireworks (Firepass)
- Fireworks AI
- Friendli
- FrogBot
- GitHub Copilot
- GitHub Models
- GitLab Duo
- GMI Cloud
- Google Vertex
- Google Vertex (Anthropic)
- Groq
- Helicone
- HPC-AI
- Hugging Face
- iFlow
- Inception
- Inference
- IO.NET
- Jiekou.AI
- Kilo Gateway
- Kimi For Coding
- KUAE Cloud Coding Plan
- Lilac
- Llama
- LLM Gateway
- LMStudio
- LucidQuery AI
- Meganova
- MiniMax (minimax.io)
- MiniMax (minimaxi.com)
- MiniMax Token Plan (minimax.io)
- MiniMax Token Plan (minimaxi.com)
- Mistral
- Mixlayer
- Moark
- ModelScope
- Moonshot AI
- Moonshot AI (China)
- Morph
- NanoGPT
- NEAR AI Cloud
- Nebius Token Factory
- Neuralwatt
- Nova
- NovitaAI
- Nvidia
- Ollama Cloud
- OpenAI
- OpenCode Go
- OpenCode Zen
- OpenRouter
- OrcaRouter
- OVHcloud AI Endpoints
- Perplexity
- Perplexity Agent
- Poe
- Privatemode AI
- QiHang
- Qiniu
- Regolo AI
- Requesty
- routing.run
- SAP AI Core
- Sarvam AI
- Scaleway
- SiliconFlow
- SiliconFlow (China)
- STACKIT
- StepFun
- submodel
- Synthetic
- Tencent Coding Plan (China)
- Tencent TokenHub
- The Grid AI
- Together AI
- Umans AI Coding Plan
- Upstage
- v0
- Venice AI
- Vercel AI Gateway
- Vivgrid
- Vultr
- Wafer
- Weights & Biases
- xAI
- Xiaomi
- Xiaomi Token Plan (China)
- Xiaomi Token Plan (Europe)
- Xiaomi Token Plan (Singapore)
- Xpersona
- Z.AI
- Z.AI Coding Plan
- ZenMux
- Zhipu AI
- Zhipu AI Coding Plan
ct_live_* API key. For control-plane management (workspaces, namespaces, actors, keys, billing), CleverThis exposes a public A2A JSON-RPC surface that the webapp uses itself; any A2A-compliant client, the CleverAgents CLI, or your own HTTP client can drive it with full parity.
CleverThis is a sustainable AI gateway with hosted Actor endpoints — by the team behind CleverThis.
Join the newsletter
Product updates and engineering notes. No spam, ever.