Free, and that is the end of it
MIT licensed and public on GitHub. No account, no tier, nothing that turns out to cost money later. Where a tool can do more with a paid key, it says which part and keeps working without one.
gledach / about
About
gledach is a shelf,
not a platform.
Eight free tools, written by the same person, that do not pretend to be one product. Each takes whatever shape suits it: a command you type, a dashboard you open, an API an assistant queries, or just writing. What they share is the promise, not the plumbing.
gledati (slav.): to watch. What a good analyst does before speaking.
What this is
Free, small, and built out of things that already exist.
The home page lists six rules about what every tool promises you. These five are about how they are made, which is the part people usually want to know before they trust one.
As few moving parts as possible
Five of the eight are Node programs with nothing to install beyond Node itself: whenrun, phonehome, signal-patterns, talk and snare. leash is a single Windows exe with no runtime to install. recall ships its own CSV and XML readers rather than pull in two dependencies for forty lines of work. signal needs three runtime dependencies, and that is the heaviest thing on the shelf.
Do not reinvent what already exists
The ones pointed at the outside world read public registers, public feeds and public APIs that were already published. talk reads the TCP table on the machine it runs on. snare reads the hook files already on that machine. leash reads the volume change log and the common persistence points already on that machine. None of it is scraped out of somewhere it should not be, and none of it is a database somebody rebuilt. The work is reading what is there properly and saying how sure it is.
Made for an assistant as much as a person
Four of the eight answer an AI assistant directly over MCP: signal, whenrun, phonehome and recall. Each result is structured data with a freshness flag attached, so the assistant can tell a quiet week from a broken collector. talk, snare and leash record on the machine and do not speak MCP. signal-patterns is the writing.
Small enough to copy
Each one is a worked example of a single idea, kept readable on purpose. If you want to build something like this rather than use it, signal-patterns is the written version and every tool here is the code that goes with it.
What they are for
Four questions, eight tools, nothing to buy.
Every tool here started as something somebody was checking by hand. If one of the questions below is yours, the tool that answers it is free, runs on your own machine, and takes about fifteen minutes to set up.
Who is moving in my market, and what did I miss?
signal watches eleven public sources around the companies you name: news and Reddit feeds, Hacker News, GitHub releases, YouTube captions, certificate logs, sitemap changes, search interest and more. It scores what it finds, groups the things several publishers corroborate, and answers your AI assistant over MCP. Three of those sources see a move weeks before anyone announces it.
Useful if you do competitive intelligence, product marketing, sales enablement or investing, and you keep finding out late. signal-patterns is the written version if you would rather build your own.
When should this job run so it costs less?
whenrun reads day-ahead electricity prices and the renewable-share forecast, works out the cheapest or greenest window that still meets your deadline, and starts the job there. Backups, builds, an EV charger, a heat pump.
German power can cost three times more in the evening than at midday, and since 2025 every supplier there has had to offer a dynamic tariff. This acts on that without tying you to one supplier's app.
What is my own stuff doing, and is any of it recalled?
phonehome records which domains every device on your network contacts and tells you when one starts talking somewhere new. It blocks nothing and uploads nothing: the job is finding out, which is different from filtering.
recall checks four public registers against a list of the things you actually own: EU product alerts, German medicine shortages, US vehicle campaigns, and software past its security support date. Every match is graded, and a register it could not reach is reported as unreachable rather than clean.
Which API did my coding agent hit, and is it leaking sockets?
talk samples TCP connections on Windows, follows each process up to its parent, and reads the certificate name of a new public host. A long session can open connections and never close them. When the client port range fills, other programs fail in a way that looks like a dead network. talk leaves a record of who, which host, and whether the count is climbing. It does not close a socket.
snare remembers the hooks a coding agent runs at startup, and the editor tasks that run when a folder opens. It prints only when that set changes, and it does not delete a hook.
leash takes a snapshot of the machine before a session and another after, then shows what changed outside that repo: files, startup entries, services, packages and ports. It never deletes, kills or blocks anything.
Eight of twelve slots are filled. A tool goes up when it runs on someone else's machine rather than only on mine, and the empty slots stay empty until one does. The six rules every tool follows are on the home page.
Seeing it first
Anything worth looking at gets its own subdomain.
You should be able to watch a tool work before you decide to install it. The convention is one address per tool, replaced in place.
<tool>.gledach.de signal.gledach.de a real deployment, exported to one static file
Real data, not a mock
The export asks the running viewer the same questions a browser asks and inlines the answers, so a preview cannot drift from the thing it was taken from.
Dated and read only
A snapshot, not a live instance. Nothing you click reaches a database, and every number is true as of the export.
One per tool, replaced in place
A re-export overwrites the preview, so you are looking at the most recent snapshot rather than an archive of all of them.
Generated text is labelled
Battlecards, briefs and convergences are model output, not verified fact. They are a starting point for a person, which is how the tool itself describes them.
Asking for a tool
The queue is public, including what was turned down.
Is there something you keep checking by hand that a tool could watch for you? That is the whole question, and it is worth asking even if the answer turns out to be no.
Three questions decide it
What you check by hand and how often. Where it is published. Whether seeing it needs an account. Those three answers decide whether a thing is watchable at all, so the form asks exactly them and nothing else.
Built in order of who it helps
A request that would help more people than the person asking is built first. The open issues are how you check that claim rather than take it on trust, and rejected requests stay visible with the reason attached.
In the open, not in an inbox
There is no request address to write to. A request sent privately is invisible to everyone else, cannot be labelled, linked or voted on, and quietly becomes a list only one person can see. The queue is the whole list precisely because there is nowhere else for one to go.
Who makes it
One person, in the open.
Everything here is MIT licensed and public on GitHub, including the source of this website. There is no company behind it, no funding and nothing to sign up for.
Creator
Aleksandar Perisic
On LinkedIn, which is the best place to reach a person rather than a queue.
LinkedIn →
Organisation
github.com/gledach
Every repository, including this site and the public request queue.
GitHub →
Updates
gledach on LinkedIn
How you hear when a slot fills, which is the only announcement channel there is.
Follow →