Loki — the global agent

yggdrasil.loki is one agent that adapts to wherever it runs: it detects the backends it can reach (offline, never raises), acts as a token / credential provider (chiefly Databricks), and dispatches pluggable skills. Drive it from code (Loki) or the terminal (ygg loki).

For the full walkthrough — capability detection, the token provider, skills, engines, tools, and the CLI — see the Loki guide. Full signatures live in the auto-generated API reference.

from yggdrasil.loki import Loki

loki = Loki.current()
loki.card()              # identity + reachable backends + skills
loki.databricks          # the live DatabricksClient, or None
loki.run("genie", question="revenue by region last month")

Surface

Symbol	Role
Loki	The agent — current(), card(), run(), databricks, backends()
Backend / detect	Offline capability detection (databricks, node, local)
LokiSkill / register	Discoverable, backend-aware capabilities + the registration decorator
TokenEngine / Completion	The reasoning-engine contract (Claude, OpenAI, Ollama, transformers, local)
Tool / Toolbox / filesystem_toolbox	Callable tools exposed to an engine
LokiSession / LokiMemory	Conversation session + persistent memory
TokenMeter / price_for	Token accounting + model pricing (METER)

Register a skill

from yggdrasil.loki import LokiSkill, register

@register
class Hello(LokiSkill):
    name = "hello"
    description = "say hi"
    def run(self, agent, *, who="world", **_):
        return f"hello {who} from {agent.user}@{agent.host}"

CLI

ygg loki is a thin shell over Loki — status / capabilities / skills / engines / tools / reason / do / token / run.