Agents that survive
crashes and pauses.
An async-first runtime where agents call tools, remember across sessions, pause for human approval, and resume after a process restart — without you writing any of that infrastructure. Durable execution, event-sourced, like Temporal — scoped to LLM agents.
Not just a prompt wrapper. A resilient runtime engine for real-world agents.
Sandboxed tool chaining, autonomous subagent swarms, long-term memory compaction, security guardrails, and dynamic code interpretation.
Built for mission-critical operations. Run agents with complete safety, budget control, and reliability.
Automatic crash recovery, hard spend ceilings, zero-cost pauses for human approvals, and full step-by-step activity visibility.
Layered for modularity and isolation. From pure domain contracts to production serving.
A 4-tier architectural hierarchy spanning 13 isolated concepts: types → telemetry → tools → models → stores → documents → workspace → safety → context → middleware → runtime → agents → testing.
Runtime & Middleware
The durable execution engine: process crash recovery, append-only WAL journals, lifecycle interceptors at TURN, CHAT, and TOOL stages, PII masking, rate limiting, and sliding-window token compaction.
# Tier 3: Durable Runtime Engine & Interceptor Middleware
from substrate.runtime import Runtime, RunOutcome
from substrate.middleware import MiddlewarePipeline, PIIDetectionMiddleware, RateLimiterMiddleware
from substrate.context import ContextConfig, SlidingWindowCompaction
middleware = MiddlewarePipeline([
PIIDetectionMiddleware(mask=True),
RateLimiterMiddleware(max_rpm=60),
])
async with Runtime.open("./.substrate", middleware=middleware) as runtime:
outcome = await runtime.run(agent, "Generate monthly report")
print(f"Status: {outcome.status.value}")
# Inspect immutable execution trace
for event in await runtime.read(outcome.run_id):
print(f" {event.kind}: {event.payload}")Open source at the core. Simple plans in the cloud.
Run it yourself for free, start on the hosted Free plan, bring your own model key for more, or partner with us for an enterprise deployment.
Run Locally (Free)
100% open source under MIT. Deploy on your laptop, Docker, or Kubernetes with zero platform limits.
Start on the Hosted Plan
A few messages a day on our models, with documents, memory and scheduled tasks. No credit card needed.
Explorer
Connect your own model key for a much higher daily limit. A flat monthly fee covers our processing; no markup on model costs.
Consulting & Sales
Need custom agent architecture, private VPC deployment, or bespoke tools? Work directly with our engineering team.