The open‑source
agent runtime platform.

Agentic runtimes that are sovereign, sandboxed, and multi-tenant.

$ curl -fsSL https://get.appstrate.dev | bash

Features.

A real agent runtime + the platform that schedules and runs it.

Built on pi, the open‑source agent orchestrator

The prompt → tool → skill loop behind every agent. read · write · edit · bash built-in, plus your TypeScript extensions and markdown skills. Stateful sessions, streaming events.

Memory & state

Typed state per run, persistent context per user, replayable history.

Connect to anything

Gmail, Slack, ClickUp, Notion, Stripe, or any internal API. Any auth method, handled.

Cron scheduling

Distributed cron. Exactly-once, timezone-aware.

Inline runs

No setup. POST an inline manifest + prompt → get a run.

Driveable from any coding agent

Claude Code, pi CLI, Cursor. Drive Appstrate from any agent.

Drive Appstrate from
any coding agent.

Drive Appstrate from
the web.

Runtime agents are the new standard.
Appstrate makes them secure and sovereign.

Claude Code, OpenClaw, Antigravity proved it: agents with filesystem, tools and memory are 100× more capable than chatbots. But they run locally, one instance per dev.

  • Security: credentials stay hidden. Sidecar proxy means the LLM never sees tokens. Chain agents infinitely, same isolation.
  • Collaboration: local runtimes are one dev, one agent. Appstrate deploys agents centrally and controls access per role: one user can drive dozens of agents, each run isolated, audited, and billed separately.
  • Scale: your laptop doesn't. Claude Code chokes at 5+ tabs. Appstrate can run 100+ agents in parallel.

Docker for processes. Stripe for payments. Lambda for functions. Appstrate for agents.

One agent, one package

Your agent's configuration lives in code, not in a UI builder.
Versioned, reviewable, portable to any Appstrate host.

Built on the open AFPS standard
my-support-agent/
├── manifest.json config + dependencies
└── prompt.md system prompt
// manifest.json
{
"$schema""https://afps.appstrate.dev/packages/schema/v1/agent.schema.json",
"name": "my-support-agent",
"version": "1.2.0",
"type": "agent",
"dependencies": { // npm-style
"@appstrate/skill-classify-tickets": "^2.0.0",
"@appstrate/tool-stripe": "^1.4.0",
"@appstrate/provider-gmail": "^3.1.0",
"my-org/custom-refund": "~1.0.0" // private scope
}
}
Code-first, open specification
Prompt
prompt.md
agent
@my-org/support
v1.2.0
dependencydependencydependency
Skills
×3
classify-tickets
Tools
×3
stripe
Providers
×3
gmail
Portable across hosts
Semver versioned
Git reviewable
Rollback-safe
CISO-auditable
Integrity SHA-256
Enterprise
Personal
Deterministic
Autonomous
runtime agents
Workflow Builders
Managed Runtimes
Coding Agents
Personal Agents
Appstrate
Appstrate

A unique position in the market.

Autonomous unlike workflow builders, sovereign unlike managed runtimes, isolated unlike coding agents, team-native unlike personal agents.

See the full comparison →

Self-host free.
Cloud when you need it.

Infrastructure and LLM tokens are billed separately. You choose.

Infrastructure · the runs
Free forever
Self-host

The full Appstrate. No feature gates. No rug pull. Apache 2.0, yours forever.

  • Unlimited agents, runs, end-users
  • Your hardware, your data
  • Full Appstrate feature set
Cloud
Coming soon

Hosted by Appstrate. Zero setup. Production scale, SSO, audit, SLA.

LLM tokens · optional, your choice
Bring your own key
BYOK
pay your LLM directly

Claude, GPT, Gemini, Mistral, or any OpenAI-compatible local model (Ollama, vLLM, LM Studio). Appstrate charges nothing for tokens.

Appstrate Credits
Coming soon

Prepaid credits, one invoice across all models. Passthrough + small margin.

The bolt. The brackets.

Appstrate mark
The bolt

The agent.

Autonomous, goal-driven, powered by the LLM of your choice. It reasons, calls tools, returns structured output.

[ ]The brackets

The sandbox.

Docker isolation per run. Sidecar proxy for credentials. RBAC, audit logs, rate limits.