Claude models via API — AGNT's primary LLM provider for intent classification and response generation.
Anthropic API + AGNT
Claude models via API — AGNT's primary LLM provider for intent classification and response generation.
Official sourceWhat it is
The Anthropic API provides direct access to Claude models (Haiku, Sonnet, Opus) with structured tool calling, streaming, and system prompt control. It is the most capable API for agentic workloads that require multi-step reasoning with tool use.
Key differentiators for agent builders: native tool_use blocks with structured input/output, a clean system/user/assistant message format, and the extended thinking feature that lets the model reason through complex queries before committing to a response.
Where AGNT fits
- AGNT's `llm_gateway.py` uses the Anthropic API as its primary provider. Every user conversation routes through Claude — Haiku for free/starter tiers, Sonnet for pro tier — with tool calling for venue search, bookings, calorie scanning, and dupe search.
- The $500/day global spend cap and 30-concurrency semaphore in AGNT wrap all Anthropic API calls uniformly. Builders composing with AGNT inherit these safety rails without implementing their own rate limiting.
- AGNT's agent hiring flow uses the Anthropic API to power Paperclip's prompt compilation — every new fleet agent gets a compiled system prompt generated via Claude before it enters the runtime.
Integration recipes
Prompts & playbooks
Links
Share as social post
AGNT + Anthropic API: Claude models via API — AGNT's primary LLM provider for intent classification and response generation. https://agntdot.com/stack/anthropic-api
165 / 280 chars