Agents

Anky exposes three agent tiers. Each tier wraps the same underlying writing/guidance/story system but with different levels of autonomy, memory, and orchestration.

Pick the tier that matches your use case. Copy the endpoints. Do weird stuff.

The Three Tiers

Tier	What It Does	Memory	Orchestration	Cost
Dumb	Single-shot write + feedback. Stateless.	None	None	Free
Smart	Session-aware orchestration. Chains writing → guidance → stories.	Session-scoped	Multi-step	USDC per session
AGI	Autonomous loop with persistent memory, adaptation, and self-directed writing prompts.	Persistent cross-session	Full autonomous	USDC per engagement

Design Philosophy

The tiers are not about intelligence. Ollama and Claude are both "smart" in the traditional sense. The tiers are about autonomy:

Dumb: You control everything. The agent is a function call.
Smart: You start a session. The agent handles the chain.
AGI: You engage. The agent drives the practice.

The progression mirrors the writing practice itself: at first you need a prompt ("write about X"). Then you need a container ("here's your session, I'll handle the rest"). Eventually you just sit down and write because the practice has become self-sustaining.

Authentication

All agent endpoints require a bearer session token from the v2 seed auth flow. See Authentication.

bash

Authorization: Bearer your-session-token

Base URL

https://anky.app/agents/{tier}/

Where {tier} is dumb, smart, or agi.