RunLayer
Open Console

The execution layer
for teams and agents

Dispatch runs. Route to metal GPU nodes when available, fall back honestly, and keep an immutable cryptographic receipt.

Open consoleAPI reference
NewTransparent Routing Protocol v0.1
dispatch_run.json•POST /v1/runs
Request Configuration
1234567891011121314
curl -X POST "https://api.runlayer.dev/v1/runs" \
-H "Authorization: Bearer rl_sk_live_9f82..." \
-d '{
"agent_id": "wallet-audit",
"model": "llama-3.3-70b",
"routing": {
"policy": "metal_prefer",
"timeout_ms": 8000
},
"input": {
"target_address": "0x7a250d5630B4cF539...",
"deep_scan": true
}
}'
Workspace:ws_production_relay
Hardware Ready
Execution Response
200 Completed
Route:network:node_us_east_4090
Hardware Metal
Latency: 1,140 ms
•
Usage: 1,420 tokens
•
Speed: 82 tok/s
{
"status": "cleared",
"risk_score": 0.038,
"findings": [
"✓ Bytecode provenance verified",
"✓ Multisig threshold valid (3-of-5)"
],
"execution_provenance": "rtx_4090_austin_node"
}
Cryptographic Run Receipt
receipt_id: rcpt_01J8XK9A7B_metal_verified
hash: sha256(input + agent + output + node_4090)

Integrates natively with leading AI models & agent frameworks

OpenAI
Vercel AI SDK
Anthropic
C
CURSOR
O
Ollama
LangChain
GROQ
🤗Hugging Face

A new species of execution layer.

Purpose-built for modern agentic workloads with transparent routing at its core, RunLayer sets a new standard for metal execution, honest fallback, and verifiable receipts.

FIG 0.1

Purpose-built relay

RunLayer is shaped by the real reliability demands of agent developers: fast router dispatch, honest fallback, and zero silent degradation.

FIG 0.2

Powered by metal

Connect decentralized provider machines directly. Verified RTX 4090s and H100 clusters execute inference over authenticated WebSocket channels.

FIG 0.3

Designed for speed

Sub-millisecond route resolution with verifiable cryptographic receipts. Generates auditable SHA-256 proofs without adding execution latency.

The full execution lifecycle

RunLayer is the relay from agent definition to verified completion — dispatch runs, route to community metal when it's online, fall back with disclosure, and keep a receipt on every finish.

Platform overview

Configure agents

Define reusable agents in a workspace — model, system prompt, tools, and optional memory. Config only; nothing runs until you dispatch.

View docs

Dispatch runs

Your app or the console sends a run with agent_id and input. The orchestrator authenticates, checks caps, and queues execution.

View docs

Honest routing

The router assigns online GPU nodes when model and region match. Otherwise a disclosed fallback completes the run — never silent.

View docs

Receipt & output

Every completed run returns output, token usage, receipt_id, and routing metadata so integrators can audit what actually executed.

View docs

First-class
developer experience

We are a team of engineers who love building tools for other engineers. Our goal is to create the agent execution platform we've always wished we had — one that just works.

Delivered
HTTP 200: { "id": "85e91b9f-9c52-4544-84c4-59056588137d" }
HTTP 200: { "id": "7d872b23-5b21-49c7-b867-56ddc2452a2b" }
HTTP 200: { "id": "8b76fb96-1ad5-4f18-8697-e9bfd70e3502" }
HTTP 200: { "id": "637ec88a-ad66-44e3-bf0c-d664a33ac54b" }
HTTP 200: { "id": "41d8e129-f14a-4a27-99e2-c01bb740e21a" }
HTTP 200: { "id": "9a22f30b-55c1-4b88-8123-ec91147a9821" }
HTTP 200: { "id": "2c4e8f1a-3b9d-4e2f-a1c5-8d7e6f5a4b3c" }
HTTP 200: { "id": "f8a3b2c1-0d9e-4f8a-b7c6-d5e4f3a2b1c0" }
HTTP 200: { "id": "85e91b9f-9c52-4544-84c4-59056588137d" }
HTTP 200: { "id": "7d872b23-5b21-49c7-b867-56ddc2452a2b" }
HTTP 200: { "id": "8b76fb96-1ad5-4f18-8697-e9bfd70e3502" }
HTTP 200: { "id": "637ec88a-ad66-44e3-bf0c-d664a33ac54b" }
HTTP 200: { "id": "41d8e129-f14a-4a27-99e2-c01bb740e21a" }
HTTP 200: { "id": "9a22f30b-55c1-4b88-8123-ec91147a9821" }
HTTP 200: { "id": "2c4e8f1a-3b9d-4e2f-a1c5-8d7e6f5a4b3c" }
HTTP 200: { "id": "f8a3b2c1-0d9e-4f8a-b7c6-d5e4f3a2b1c0" }

Test mode

Simulate agent runs and exercise the routing API without dispatching untracked workloads to production metal.

Learn more
CompletedSep 29 02:06:25
fromsandbox@runlayer.devonagent wallet-auditon metalnode_us_east_4090regionus-east-1
DisclosedSep 29 02:06:18
tomorning-briefwith routingfallback:groq
CompletedSep 29 02:06:09
fromapi@runlayer.devonagent token-scoutreceiptrcpt_01J8XL9C2F
DisclosedSep 29 02:05:52
todeep-researchwith routingfallback:groq
CompletedSep 29 02:06:25
fromsandbox@runlayer.devonagent wallet-auditon metalnode_us_east_4090regionus-east-1
DisclosedSep 29 02:06:18
tomorning-briefwith routingfallback:groq
CompletedSep 29 02:06:09
fromapi@runlayer.devonagent token-scoutreceiptrcpt_01J8XL9C2F
DisclosedSep 29 02:05:52
todeep-researchwith routingfallback:groq

Modular webhooks

Receive real-time notifications when runs complete, route to metal, fall back with disclosure, or issue a receipt.

Learn more

Execute with proof,
not black-box inference

  • Disclosed routing

    Every finished run includes routing metadata — network node id, region, and latency, or a labeled fallback. No silent off-network execution.

  • Receipt on completion

    Integrators get a stable receipt_id with usage and routing source for support, billing disputes, and internal audit trails.

  • Workspace API keys

    Issue scoped keys (rl_sk_…) per project. Your backend calls RunLayer; end users never need to see the console.

  • Agent definitions

    Store model tier, system prompt, tool policy, and optional memory as config — materialized into a run only when you dispatch.

  • Community GPU nodes

    Operators run runlayer-node on metal. When a node matches model and region, the orchestrator assigns work over WebSocket.

  • Honest fallback lane

    If no suitable node is online, configured adapters complete the run with disclosed fallback in the API and console.

  • Runs-first API

    POST /v1/runs and GET /v1/runs/:id are the core contract. Playground chat is one preset agent — production stays API-driven.

  • Caps and metering

    Track prompt and completion tokens per run. Spend caps per workspace and API key land in v1; design for limits from day one.

  • Integrator-ready billing path

    Stablecoin settlement, x402 pay-per-run, and optional on-chain receipt anchors are on the roadmap — not the v0 launch blocker.

Everything in your control

The console is built for builders — inspect runs, read routing truth, manage agents and keys, and debug production workloads without treating chat as the product.

Every run in one place

Filter by agent, status, and routing type. Open a run to see input, output, token usage, receipt_id, and whether execution used network metal or disclosed fallback.

RunLayer console · Run operations
RunLayer console — runs, agents, playground, and receipts