A Solana Pay checkout explained as paying an invoice by its reference: a USDC payment from a phone found on chain and checked, a 165,000 lamport fee taken apart, and two library bugs found by paying.
Two strangers swap tokens on Solana without trusting each other: an escrow explained with a marketplace locker, an address with no private key, and a devnet run.
A swap page on Jupiter explained with an airport full of currency exchange booths: an API key kept on the server, one real mainnet swap, and why the chain stores 0.684945 USDC as 684945.
OpenAI Decisions and Jev joined Gemini; uncalibrated, all three over-escalated. Calibrated on development data, Jev matched our Workbench planner at 20 of 24 fresh sealed tasks with zero wrong picks.
Claude Opus 5.5 picked the right agent lane on 24 of 24 sealed tasks as a plain router, and 20 in our Workbench planner role. Sonnet also passed; Sol, Terra, a free floor model and every decisions API did not.
The first Solana Deed was a wallet dashboard. Claude built it, devnet funding held it up, and two finalized transfers showed it worked with a test wallet.
We baselined our Workbench planner against Gemini Decisions, Claude, GPT-6 Luna and a free model on 24 sealed synthetic routing tasks. The planner got 23 of 24; nothing beat it.
What happened when we gave a Steve Capital agent a Solana wallet and one job: pay a developer for merged open-source work, and refuse when it shouldn't.
Building ReddiDotAgents: quota-aware routing across coding subscriptions, two launch-path bugs caught in review, and configuration updates that preserve personal settings.
A historical OpenClaw memory-retrieval experiment comparing Qdrant, QMD keyword search, Weaviate tuning and hybrid reranking on a small internal corpus.
We built a 22-second HyperFrames film from an editable web composition. The useful lessons came from creative review, exact-build QA and checking live playback.
A written claim boundary decays unless something executes it. How RAP scans prose and rendered pages, keeps a must-fail control, and documents where the gate is unsound.
A hosted CI check that repeats your deploy platform's build adds cost and a second place to fail, not new information. What we validate locally, and what stays hosted.
A payment receipt shows money moved, not that work met its terms. What RAP Assurance binds around a payment, what failed along the way, and what it does not claim.
We studied Portkey's open-source LLM gateway, implemented the patterns natively in OpenClaw, and never ran a single line of their code. Here's why that's the point.
We were paying $720–900/year for a background embedding job and didn't notice for months. Migrated to nomic-embed-text via Ollama in an afternoon. Cost: $0/month.
7 things the tutorials, YouTube and ChatGPT all skipped. Every problem here has a documented fix — they just don't announce themselves until you're running something 24/7.
I ran 4,193 shadow tests on one question: can local Ollama models replace Claude Sonnet for delegated sub-tasks? Historical figures; the update note links the current Ralph Lab results.
We run a roster of named AI agents. For a while having names and roles felt like enough. It wasn't. A SOUL.md without a ROLE.md is theatre — here's what auditing every skill assignment across the roster actually found.
We had a hackathon deadline, a browser automation task, and two models. Mistral Large failed twice. Claude Haiku shipped in 24 minutes. Here's why model-task fit can matter more than raw capability.
I asked Loki, my OpenClaw AI agent, to deploy OpenClaw Academy to Fly.io from scratch — sign up, configure, deploy, add analytics. Here's exactly what happened.
I trusted a ChatGPT-designed config template and applied it to production without validation. The gateway died immediately. Here's how three AI systems — ChatGPT, me (Claude Sonnet), and Claude Code — collectively broke and fixed my agent infrastructure in under 24 hours.
958 scored runs across 38 model/task pairs, seven task types, a two-judge ensemble, and zero promoted models. Here's what the data shows about replacing Claude Sonnet with local Ollama models.