Cognitio CLI · Falcon

What it is #

cognitio is the terminal client for the Response series. Open it inside a repository and it starts a session: a banner, a prompt, and Cognitio Infer working in that directory. Describe a task; it looks around with list_dir, read_file and search, proposes changes with write_file, checks them with run, and reports. Reads are free. Every write shows you the diff and every command is printed, and each waits for your y before anything happens. The conversation carries from one task to the next.

It talks to the Falcon API with your preview key and nothing else: no model runs on your machine, no account is created, and nothing leaves your computer except the request you make. The same binary also answers one question, keeps a plain chat with Cognitio, and summarises a file.

The look is made for a light terminal: ink-grey text on your own background, one blue accent for the prompt and the answers, orange bullets for tool calls with their output folded beneath, a diff in green and red before every write, a spinner while the model works, and a status line with steps, tokens and seconds after each task.

Install #

bash

npm install -g https://falconlab.app/downloads/cognitio-0.2.3.tgz
cognitio config --key <the key>
cognitio

The package is a single tarball served from this site, ten files and no dependencies, for Node 20 or later. It is not on a public package registry during the private preview. Your key comes from the portal; cognitio config stores it in your home directory, readable by you alone.

What it does #

Command	What you get
`cognitio`	A session: Cognitio Infer working in the current repository, asking before every change, remembering the conversation
`cognitio "a question"`	One answer from Cognitio Infer, reasoning first, at low thinking by default
`cognitio chat`	A plain conversation that remembers the last eight turns; `/quit` to leave
`cognitio summarize notes.md`	Cognitio's four-to-six-sentence summary of a file, or of whatever you pipe in
`cognitio code "a task"`	One task through the harness, then exit; for scripts
`cognitio config`	Your key, API address, default model and thinking effort

In a session, /yes turns confirmations off for a repository you can throw away, /think and /model change what answers, /reasoning shows the thinking before each step, /clear forgets the conversation and /cost shows what the session has used.

A code session #

text

❯ greet.mjs ignores its name argument; make it print hello, <name> and run it to check
  ✻ thought
● list_dir .
  ⎿  README.md
     greet.mjs
● read_file greet.mjs
  ⎿  const name = process.argv[2] || "world";
     console.log("hello");
◆ The script reads the name but never prints it. I will use it in the greeting and run the file.
● write_file greet.mjs (72 characters)
  greet.mjs
  - console.log("hello");
  + console.log(`hello, ${name}`);
? Replace greet.mjs (72 characters)? [y/N] y
  ⎿  wrote greet.mjs (72 characters)
● run $ node greet.mjs Ada
? Run: node greet.mjs Ada [y/N] y
  ⎿  exit 0
     hello, Ada
◆ greet.mjs now prints the name given on the command line; node greet.mjs Ada printed hello, Ada.
✓ 5 steps · 2,452 tokens · 31.0 s · cognitio-infer low

Every path stays inside the repository, a declined write or command is reported back to the model so it can change course, a call repeated after the same failure is stopped by the harness, and the loop ends after thirty steps. --yes, or /yes in a session, skips the questions for a repository you can throw away.

How it is metered #

Each answer, each chat turn, each summary and each step of a code session is one request in the response bucket of your key, the same bucket the chat site and the API's chat routes use. cognitio config shows the key it is using; the portal shows what is left this month. Keys, quotas and the routes behind the commands (/v1/chat, /v1/summarize, /v1/agent) are described on Access and API conventions.

Questions #

Does it run a model on my computer? #

No. Every command is a request to the Falcon API; the models run on Falcon's GPU. Without a network connection and a key it does nothing.

What can the code mode change? #

Only files inside the directory you run it in, and only after you answer y to the diff it shows. Commands run in that directory with your permissions, again only after you confirm them. It never runs anything on its own unless you pass --yes or turn /yes on.

Which model should I use? #

Cognitio Infer for anything that has to be worked out or that uses tools; Cognitio for a quick answer or casual chat. The session and cognitio code use Cognitio Infer unless you switch with /model.

Where is the full reference? #

The Command line page in the documentation lists every command, flag, tool and error message, and the tarball's README repeats the essentials.