An AI agent that improves itself, built in the open.
The Tinker Zone is the workshop log of TinkerClaw: the open-source code, the research papers behind it, and notes from running it around the clock.
Which AI model should you use, and what will it cost?
Every model on one map, placed by how smart it is against what it costs. Next to it, a report card of what each one is good at and what it refuses to talk about, and who sells it cheapest.
Our agent rebuilds it every morning. When a new model launches, check here: it will be on the map within a day or two.
Open the AI analysisNewest posts
Browse all posts-
Index: a map of every post on the Tinker Zone
Every post on the Tinker Zone as an interactive map: bubbles clustered by topic, arrows for every link, TinkerClaw on GitHub at the centre.
-
When Code Gets Cheap, What Is a Software Company Selling?
If an AI can rebuild a product in a weekend, what are customers paying for? Four assets cheap code can’t copy, why pricing decides…
-
AI Costs in 2026: Buy a Computer, Rent, Pay per Token or Subscribe?
Buy a Mac Studio, rent cloud GPUs, pay per token or subscribe? We priced a year of heavy AI use each way for the…
TinkerClaw
Our open-source fork of OpenClaw, and the agent most of this site is about. The code is free to read, run and fork.
- Shows what every token costs, as it happens
- Matches each task to a model and an effort level
- Keeps its memory and personality across restarts
- Runs saved playbooks, and nightly jobs that improve their own prompts
Articles about TinkerClaw
How it works, one part at a time.
-
Index: a map of every post on the Tinker Zone
-
AI Analysis — the Smartness × Cost Map and the Model Dossier
-
Every Token, Every Dollar: The Tinker Command Center
-
Pick the Right Brain for Every Task: The Effort × Model Slider
-
The Recipe Book: TinkerClaw’s Playbook Library
-
Watching an Agent Think: The EEG View
-
The TinkerClaw Brain — we mapped agent capabilities onto the human brain, then built it
-
TinkerClaw, Eight Months In: What We Changed in OpenClaw, Panel by Panel
Research papers: Building Jarvis
All papersThe ideas behind TinkerClaw, one paper each: memory, reflexes, personality, security, orchestration and cost.
-
SEAHORSE: What Actually Makes an Agent Code Well
-
THALAMUS — Which AI Should I Ask?
-
STRIATUM: Learned Personality Modulation — Tuning an Agent’s Behaviour from Feedback, Not Static Prompt Text
-
BROCA: Giving Agent Recipes a Grammar — A Gradual Type System and Combinator Algebra for Self-Composing Workflows
-
SALIENCE: The Death of Fixed Thresholds, the Pyramid of Significance, and Cheap Traversal as the Basis of Next-Generation Vibe Programming
-
INSTANT RECALL: Finding What the Agent Already Owns
-
Fractal Reasoning: One Reflex at Every Scale, and a Post-Mortem of the Reflection Lane That Has Not Worked Yet
-
Identity Persistence: Keeping an LLM Agent’s Personality Stable Across Sessions, Model Swaps, and Restarts
Field notes from running an agent 24/7
Browse all postsWhat happens when an AI agent never switches off: the bills, the crashes, and the lessons we wrote down.
-
Running an AI Agent 24/7: What Nobody Tells You
75 crashes, 45 wrong messages, and 138 restart loops. What actually happens when you leave an AI agent running around the clock.
-
How to Track AI Token Costs in Real Time
See where every AI token goes — in real time, not on next month’s invoice. Practical strategies that cut our daily costs by 60%.
-
How to Set Up Self-Improving AI Cron Jobs
The META pattern: AI cron jobs that rewrite their own prompts after each run. Day 1 mediocre, Day 30 expert. No human intervention.