Blog · page 2 of 2
Blog
Notes from the people building Anchor Terminal on what we're making, why, and what we learn from the research.
What we think, the argument these posts make, in order.
· From the data · 6 min read
The best harnesses for a personal agent, by our grades
Vlad Cealicu on the open-source programs people run to turn a model into a personal agent on their own machine, from OpenClaw to ZeroClaw and Hermes Agent, read against our grades and what each one does before it acts for you.
· The Log · 7 min read
The Log, issue 2: agents loose, money strange, a month's notice
Issue 2 of The Log, the week in AI for people who don't follow it every day. Test agents on real websites and Washington reacting, JPMorgan shutting outside agents out, Anthropic's revenue and Character.AI's bankruptcy, new models from Meta and Google, and shutdowns with a month's notice.
· Get found by agents · 3 min read
How to make your tool agent-ready
Agent-ready means an AI agent can find, understand, call, recover from and pay for a tool without a person in the loop. How we measure it, why only 174 of the 1,114 tools we list make it, and what gets a tool there.
· Get found by agents · 3 min read
How to let agents pay you, with x402
An agent can't sign up, add a card or copy an API key. x402 lets it pay inside the HTTP request instead. How a payment goes, who runs the standard, and the fifteen tools we list that accept it, with their prices.
· Agent errands · 9 min read
Personal agents sell errands they can't run, and I think they know it
Vlad Cealicu on why personal AI agents get tried once and dropped. An agent trying to book his Munich to London flight, the bot walls in the way, what brute force costs, the numbers on who comes back, and why he built LocalGhost as a memory layer instead.
· Get found by agents · 3 min read
How to make your site and tool ready for agents
What an AI agent looks for when it lands on a tool's site, how many of the 966 tools we list have each piece, what doing it on our own site did for us, and where llms.txt fits.
· Run it locally · 9 min read
Can you run Muse Spark Open locally? Hardware, RAM and cost
Vlad Cealicu on Meta's local pitch for Muse Spark Open (816B, 42B active), and how to run it locally. How much RAM and VRAM each version needs, whether it runs on a Mac Studio, a Minisforum MS-S1 Max, RTX 5090s, a DGX Spark or a CPU server, how many of each you'd need, what they cost and how fast they'd be, and when an API is cheaper.
· Opinion · 5 min read
Few people have seen what agents can do, so little is built for them
Vlad Cealicu on why so little of the web and so few tools are built for agents yet. Andrej Karpathy's funnel of who has seen what AI can do, numbers from 1,004 tools, 20 agents in parallel, and how agents will buy the tools they use, the way travel agencies once sold trips.
· Get found by agents · 4 min read
How to make your MCP server discoverable
Half the tools we list have an MCP server, and few of them are easy to find. What MCP is, why we think it's a clumsy protocol, when a plain API with an OpenAPI description does the same job, and the steps that get a server found.
· From the data · 9 min read
The tools personal assistants will reach for, and where they get stuck
Assistants such as Muse, ChatGPT and Claude are taking on errands. We reviewed 132 listings in 13 everyday categories. Most were built for people, and sign-in, payment and checkout are where agents stall.
· Opinion · 3 min read
Cypherpunks write code, and now agents read it
Vlad Cealicu on REA, an open-source tool that lets a coding agent take apart any program and rebuild it, which went from its first release in July to over 100,000 GitHub stars. Why it's what the cypherpunks wanted, and why the explosion of software is only starting.
· From the data · 5 min read
What machines read when they visit an AI tool directory
A week of Anchor Terminal traffic in charts. Comparisons, execution sandboxes, decision APIs and voice cloning, with the limits of what a server log can tell us.
· Opinion · 3 min read
Decision models, part two: train your own
Vlad Cealicu on the other way out of the decision model land grab. Coinbase fine-tuned Qwen3.5-9B on its own fraud data and says it beat frontier models. Eleven of the fifteen decision models we list publish weights, and why a company's own decisions are the best training data it has.
· The Log · 4 min read
The Log, issue 1: decide, declare, depart
Issue 1 of The Log, our week in review. The three biggest things for agent tools in the week of 5 October 2026. OpenAI and Microsoft ship decision models, Sierra and Meta publish the Personal Agent Protocol, and Google moves its coding agents to Antigravity. What each one changes, and what else happened.
· From the data · 6 min read
Nobody reads the terms. We read 1,268.
Quynh Tran-Thanh on the terms and privacy policies of 925 hosted agent tools. Who publishes none, which clauses turn up most, and why identity verification, payroll and banking data read worse than they should.
· Opinion · 4 min read
Most agent tools can't pass a due-diligence questionnaire
Vlad Cealicu on why most agent tools aren't fit for enterprise yet. We asked eight standard due-diligence questions of 760 hosted tools. Nine answer all eight in public, and a third answer three or fewer.
· Opinion · 4 min read
After a year building a personal agent, I mostly want it to help me remember
After a year building LocalGhost, Vlad Cealicu values memory, research and summaries most. What that experience means for the new Personal Agent Protocol.
· Opinion · 5 min read
Decision models are popping up like mushrooms after the rain
Vlad Cealicu on why fourteen decision models launched in under four weeks, what Microsoft, OpenAI, Cloudflare, Perplexity and the rest have at stake, and why everyone wants to see the decisions your agents make.
· Inside Anchor Terminal · 4 min read
Why we built Anchor Terminal
Vlad Cealicu on why he and Quynh built Anchor Terminal after CryptoCompare, how the benchmark and the agent-first site work, what letme.dev is for, and how the company makes money without selling rankings.