A replayable, approval-aware agent execution kernel.

Turns, permissioned tools, memory hooks, and traces — shared by Anima and future packs.

One Python runtime for CLI / Web / Mobile — no dual-engine drift.

Shipping agent products takes more than a prompt.

You need a stable turn state machine, permissioned tool paths, session context, and auditable event trails when things fail. Runtime packages that as a reusable kernel.

Core capabilities

⟳

Turn & Streaming

Session-scoped turns, SSE streaming, phase events, and recoverable LLM error handling.

🛡

Tools with consent

Shell / MCP: ask → approve/deny → execute → feed results back; Always-allow rules supported.

▣

Memory & Skills

Memory retrieval injection, Skill Markdown, Workspace / AGENTS.md project binding.

{ }

Trace & Headless

JSONL events, timeline APIs, and scriptable headless turns.

How products layer

Runtime kernel

Execution, permissions, memory hooks, and traces — decoupled from any specific UI.
Product packs

Anima adds personas, relationship state, feed, and clients.
Channels

App / collaboration software / private deploy — without rewriting the turn loop.

Build on Runtime with us?

Private deploy, industry skill packs, or joint R&D — contact us via the official domain email.

hello@xcreate.tech

{ "event": "tool_ask", "tool": "shell", "status": "pending" }