66 terms. Search by the docs name, the plain name, or what it does. Click a card for its connections, or open its full page.
The handful of names everything else hangs from.
The complete AI assistant: server, database, models, memory and history, treated as one living whole.
The open-source software itself. "Lich" + "D" for daemon, the Linux word for a background service.
The person in charge. Sets things up, watches what happens, says yes or no, fixes mistakes.
The project and its published documentation: vision, rules, how-tos and lore.
Servers, folders, config and the machine underneath.
The single running server process. It receives work, runs it and serves the web UI.
Everything that physically runs on the Linux machine: containers, services, mounts.
Where all configuration lives. Edit here, then run "lychd bind" to apply it.
One small config file that declares one thing you want to run, like a local model server.
The command that takes your settings and generates the actual service files on the machine.
Where durable files live: database storage, the workspace, private extensions.
The terminal commands you use to install and control LychD.
The running log of what the system is doing.
Setting up LychD on a real Linux machine and getting its first reply.
How a request becomes a job, a workflow and a result.
What you are asking for, in a structured form the system understands.
One accepted request, with a clear boundary around who asked, what is allowed, and what happened.
The official record of one job: its state (queued, running, waiting, done, failed) lives here.
Background workers that pick up queued jobs and push them forward.
Picks the exact workflow for a request and keeps it fixed for the whole job.
A reusable workflow, like "chat with tools" or "research a topic".
One exact, frozen version of a workflow. A job is locked to the version it started with.
One named action inside a workflow, like "fetch page" or "ask model".
Actually running the workflow for one job.
The map of steps a workflow moves through, with save points so it can pause and resume.
A job safely paused, waiting for a model to load, a human approval, or another agent.
After a crash or restart, the new process reads the records and decides what can safely continue.
Where the actual intelligence comes from and how it is scheduled.
Anything that can provide AI power: a local model server or a remote API.
An AI model running on your own GPU or CPU, managed by LychD.
An AI model hosted somewhere else, reached over the internet.
Finds a model that is ready right now and hands out a temporary pass to use it.
The only part allowed to change what is loaded on the GPU. It drains users first, then switches.
A temporary pass to use one model. The Stage Manager waits for all passes to be returned before switching.
A named group of local model servers you can start or stop together.
Who is allowed to do what, and what needs a human yes.
Checks who you are and whether you are allowed to do this, right at the moment of action.
The identity and permission tag that travels with every request.
Before risky actions, the job pauses and asks you. One yes approves exactly one thing.
A locked box where untrusted code or outside agents could run without touching the real system.
Before any data leaves your machine, a clean copy is made and checked byte by byte.
What the system remembers, and the philosophy of how it thinks.
The database. Anything committed here survives crashes and restarts.
What the assistant deliberately remembers for later, with where it came from.
What happened last time, and what it taught, kept so future work can learn from it.
Exactly what the model is shown for this step. If the essentials do not fit, it refuses rather than guess.
A Seed is something kept that can matter later. Recall is when it comes back and influences a new task.
Four jobs of a good mind: take in the request, test it against evidence, carry memory forward, and take responsibility for the result.
The web UI and its instruments.
The web app where you talk to the assistant and see what it is doing.
The chat window. Requests go in, answers and approval cards come back.
One request and everything that belongs to it, shown as one unit in the chat.
Where ongoing projects, open questions and decisions are tracked.
Look inside one job: what happened, what was recorded, and what is not known.
Shows the exact workflow version a job used, step by step.
Shows which models are ready, busy or loading, and lets you request a switch.
What gets built on top of the platform.
An application built on the platform, e.g. Blockworld (Minecraft), Reach (Discord, email), Companion (phone).
When several apps must start, wait, cancel or recover together as one job.
A Product is what you sell to a profession. A Deployment is one customer's installed copy.
Fifteen named areas where new abilities can be plugged in: web, voice, vision, security and more.
Reads the web safely. Each kind of action (search, fetch, download) needs its own permission.
Everything about speaking and listening.
Seeing and generating images, video and 3D, always with the source kept separate from the result.
Tests whether something actually meets its goal. Its findings are evidence, never automatic approval.
Tries alternatives in isolation without touching reality, then proposes the best for review.
Training or fine-tuning models from approved data. A human decides promotion.
Keeps track of the assistant's personality profiles and who did what.
Lets your assistant ask another independent assistant for help. Replies are treated as foreign data.
How decisions and changes are made in the project.
A written architecture decision: what was decided, why, and what future changes must respect.
The honest status page. If it is not marked Available or Partial there, do not claim it works.
Write down what is true and what the rules are first, then build it.
Changes to LychD files you distribute must stay open. Your own separate files and client work stay yours.
