Seiso: A Convention And Linter For AI-Written Docs
Agent-written documentation has a peculiar failure mode: the text is fluent, tidy, and quietly wrong. A setup guide states the version of the moment as if it were a requirement. Three pages retell the same field list, and two of...
Procoder: A Commit Gate Your AI Coder Cannot Talk Past
The weakest link in AI-assisted development is the moment the agent says “done”. Nobody is standing behind it. The tests may not have run, the formatting may be off, a merge conflict marker may be sitting in a staged file,...
PeerBridge MCP: A Control Room Where Agents Cannot Quietly Overstep
Running several coding agents at once is easy. Knowing what they are allowed to do, what they actually did, and whether the record of it is true - that is the hard part, and it is where most multi-agent setups...
PaperGraph MCP: Read Math Papers With Evidence, Not Guesses - Inside lotchuaz...
Ask a language model to explain a proof and it will happily invent dependencies, conflate similar theorems, and present all of it with total confidence. For mathematics that failure mode is fatal, because the value of a paper is exactly...
MemoryBridge: A Memory That Follows You Across Devices
You work through a hard problem with your AI on the phone during the commute; back at your desk you want to continue on the PC - and today that means scrolling history, copy-pasting, and re-explaining everything. Cloud full-sync is...
Glyd: Bit-for-Bit Compression for Model Weights and Data Lakes - Inside surya...
Every layer of the AI stack has a storage problem. Training runs checkpoint terabytes; inference packs GPUs to the brim with weights and KV caches; data lakes drown in logs and dumps that are written once and read constantly. Compression...
geo-score: Can AI Engines Cite Your Site? - Inside jianruntech/geo-score
Search is quietly splitting in two. One half ranks pages the way it always has. The other half answers: ChatGPT, Perplexity, Google AI Overviews, Gemini and Copilot retrieve passages, judge whether a source is worth quoting, and either cite you...
Gap-Trap: Rules With Teeth For Agent-Written Code
Every team running coding agents has the same quiet problem: the rules exist, but nothing enforces them. The agent writes a helper that already exists, crosses a layer boundary because the shortcut compiled, and follows a rule in your instructions...
Cortico: An Event-Stream Harness For Agents That Never Sleep
Most agent frameworks are built around a conversation: a request comes in, the model answers, the session ends. Cortico by Pal AI Lab - v0.1.5 pre-release, MIT, TypeScript on Node 22+ - is built around a different primitive: the event...
BankMCP: Your AI Now Reads Your Bank - Inside noskillish/bankmcp
The questions are simple. Has the invoice from Acme been paid? What did we spend on groceries in August? Which subscriptions am I paying for, and what do they cost per year? Yet the answers live behind bank logins, per-bank...
system-one-connector: Give Your AI Agent Judgments It Can Branch On - Inside ...
When an AI agent needs a quick judgment call, the usual move is to ask a reasoning model, read a paragraph back, and guess what it meant. “This seems fairly urgent” gives the agent nothing to put in an if...
Ollaya: Run Open Decision Models Locally, the Way Ollama Runs LLMs - Inside o...
Not every AI workload needs a generator. Classify this support ticket, score this churn risk, route this request by language, decide whether this email is urgent: these are questions with fixed answer spaces, and paying a token-by-token LLM to answer...
Lemmalog: A Datalog Engine for LLM Agent Memory - Inside JordyZomer/lemmalog
Most agent memory systems are retrieval problems waiting to be re-solved: embed everything, vector-search, hope the right chunks surface. JordyZomer/lemmalog starts from a different thesis entirely, that an agent’s memory should be a deductive database. Instead of remembering better than...
Jevgrep: Find Code by Asking What It Does - Inside dzhng/jevgrep
Coding agents spend a real slice of every unfamiliar task just finding the right files: grepping for guessed keywords, opening plausible folders, reading code that turns out to be irrelevant. dzhng/jevgrep attacks that step directly. It is a CLI, jg,...
Foremerge: Catch Intent Conflicts Before Code Conflicts - Inside naw103/forem...
Run two AI coding agents on the same repository at the same time and you will eventually meet the merge that should not have happened. Each agent works in its own copy of the code, each diff looks reasonable, Git...
Fauxnix: Bash For Your Windows Agent, No VM Required - Inside 20000419/fauxnix
Ask a coding agent to work on a Windows machine and watch what happens to its shell discipline. The model was trained on decades of bash, so it writes ls -la | grep foo and find . -name '*.ts' |...
EvoOntology: A Self-Evolving Ontology Layer for Data Agents - Inside ruc-data...
Raw data leaves its meaning implicit. Table names, column headers, file paths, and isolated observations rarely explain what a metric actually counts, which entities relate to which, or what constraints the business assumes. Agents that work over heterogeneous tables, files,...
Doop: Humans and AI Agents Designing on One Canvas - Inside kgoedecke/doop
Design tools have been quietly converging on the same realization: the next collaborator on a design file is not another human. Most products bolt an AI panel onto the side and call it done — you prompt, it generates, you...
Coucou: A Tiny Friend in Your Notch That Watches Your AI Coding Agents - Insi...
AI coding agents work in terminals while you work somewhere else, which means the moment they need you, a permission prompt, a clarifying question, a finished build, you are usually looking the other way. Louis-CFM/coucou solves that with charm and...
AnyJev: Typed Decisions With Calibrated Probabilities - Inside nokia-applied-...
If you have ever asked a language model to pick between a few options, you have probably noticed that the confidence it reports is more decoration than measurement. Reorder the options and the answer changes. Read the probability it puts...
second-brain-os: An AI-Maintained Knowledge Vault - Inside undefined-ui/secon...
Every knowledge system you have ever used has the same failure mode: it works for two weeks while you are enthusiastic, and then the filing falls behind and the pile wins. Bookmarks accumulate without being read, read-later queues become guilt...
quackd: One CLI For All Your Robots - Inside rokbenko/quackd
Every robot you can buy today already knows how to move. A walking duck balances itself with an onboard policy, a desktop arm picks with a learned skill, a wheeled base drives - and none of it needs an AI...
headcount: An Agent Company In Sixteen Departments - Inside cbrock84/headcount
Most agent libraries are piles of prompts. You install a mega-pack, your context fills with instructions you will never need this session, and when two skills disagree there is nothing in the repo that says who wins. The prompt pack...
geo-score: An Open Rubric For Whether AI Answer Engines Can Cite Your Site - ...
A strange kind of invisibility has arrived for websites: your pages can rank well on Google and still never be quoted by ChatGPT, Perplexity, Gemini, or Copilot, because answer engines do not rank results — they retrieve passages, decide which...