# Anthropic Messages API Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/anthropic-messages-api
> Markdown URL: https://aitinkerers.org/technologies/anthropic-messages-api.md
> Technology record last updated: 2026-09-18T15:13:43Z
> Generated: 2026-09-21T15:40:34Z

The primary interface for building Claude-powered applications via structured, multi-turn message arrays.

Anthropic's Messages API serves as the gateway to Claude 3.5 Sonnet and the Opus model family. It replaces the legacy Text Completions endpoint with a robust, role-based structure (user vs. assistant) that supports multi-modal inputs like images and PDFs. Developers use this API to manage context windows up to 200k tokens, implement tool use (function calling), and stream real-time responses with sub-second time-to-first-token performance.

- Official technology site: https://docs.anthropic.com/en/api/messages
- Public AI Tinkerers demos and talks: 2
- Result page: 1 of 1

## Recent Public Talks and Demos

### [pi-hydra: Mob Programming for Coding Agents](https://cologne.aitinkerers.org/talks/rsvp_WhIovGtreZ0)

pi-hydra is an open-source extension for pi.dev, a coding agent harness, that adds observer "heads". A head reviews the agent's work while the agent is still working. Each head is one markdown file with its own lens. The repo ships six as examples, but you write your own or tell the agent to write one. After every step a head makes an observation and decides: stay quiet, print a note for me, queue feedback for the next turn, steer the agent mid-run, or interrupt the run entirely. The author came to this from mob programming, one driver at the keyboard and several navigators watching the same screen. Most agent tooling points the other way and delegates work down to a worker that builds its own context and hands back a result. A head goes up instead. It sits above the driver, watches the same run, and can act on what it sees. It also does not have to rebuild context: each observation is a fork of the driver's own provider request, replayed byte-for-byte so it rides the prompt cache the driver already paid for. In the Anthropic sessions the author measured, one always-on head adds roughly 30% to total session cost, and a second head about the same again. A head sees everything the driver sees, so it inherits the driver's framing and blind spots too. It catches many problems while they are cheap, but it does not replace the independent review at the end. In the demo, an agent added a feature to an existing service while two heads watched. One caught a security hole the agent had inherited and never thought to check. The other caught the agent quietly swallowing errors in code it was writing at that moment. Both steered the fix in mid-run without derailing the original task. Three prompts, about two minutes.

- Event context: AI Tinkerers Cologne: World Cup Edition — 2026-07-15 — Cologne
- Public talk page: https://cologne.aitinkerers.org/talks/rsvp_WhIovGtreZ0

### [Domain-encoded system prompts: turning 15 years of accounting forensics into a live Claude diagnostic tool](https://montreal.aitinkerers.org/talks/rsvp_x6xfCYXZsaM)

I built a structured diagnostic engine that ingests raw ADP Workforce Now payroll exports and NetSuite GL data, routes them through a domain-encoded Claude system prompt, and returns a JSON-structured root cause analysis — including the specific fix — for recurring payroll-GL discrepancies that human reviewers consistently miss. The demo runs live against the Anthropic Messages API. The input is a real payroll pattern: employer burden accounts (CPP, EI, CNESST, RQAP) posting at exactly 0.500x expected in every quarter-end month for three years. The system prompt carries the diagnostic heuristics — check ratio before amount, look for N-period calendar patterns, separate gross wages from employer burden — that normally live inside a practitioner's head. Claude doesn't discover these rules. They're injected. The output is a structured JSON object with root_cause, confidence_pct, adp_fix_steps[], correcting_je_description, and cfo_memo.

- Event context: AI Tinkerers Montreal - March Demo Night — 2026-03-24 — Montreal
- Public talk page: https://montreal.aitinkerers.org/talks/rsvp_x6xfCYXZsaM

## Related Technologies

- [ADP Workforce Now](https://aitinkerers.org/technologies/adp-workforce-now) ([Markdown](https://aitinkerers.org/technologies/adp-workforce-now.md)) — 1 public demo
- [Claude Code](https://aitinkerers.org/technologies/claude-code) ([Markdown](https://aitinkerers.org/technologies/claude-code.md)) — 215 public demos
- [claude-sonnet-4-20250514](https://aitinkerers.org/technologies/claude-sonnet-4-20250514) ([Markdown](https://aitinkerers.org/technologies/claude-sonnet-4-20250514.md)) — 1 public demo
- [JavaScript](https://aitinkerers.org/technologies/javascript) ([Markdown](https://aitinkerers.org/technologies/javascript.md)) — 40 public demos
- [JSON](https://aitinkerers.org/technologies/json) ([Markdown](https://aitinkerers.org/technologies/json.md)) — 20 public demos
- [OpenAI Codex](https://aitinkerers.org/technologies/openai-codex) ([Markdown](https://aitinkerers.org/technologies/openai-codex.md)) — 12 public demos
- [pi](https://aitinkerers.org/technologies/pi) ([Markdown](https://aitinkerers.org/technologies/pi.md)) — 2 public demos
- [TypeScript](https://aitinkerers.org/technologies/typescript) ([Markdown](https://aitinkerers.org/technologies/typescript.md)) — 205 public demos