FAQ
About C5 #
Do I need to be a programmer? #
No. You ask for things in plain words. Being comfortable with a terminal helps for install and updates, but the app itself is a normal website in your browser.
Do I need internet? #
Only for a few things: installing, updating, activating, and using SI models hosted by other companies.
C5 also downloads a few models once, from Hugging Face, for features that run on your own computer:
| Feature | Download |
|---|---|
| Memory search (semantic search) | about 2.26 GB |
| Offline voice | about 60 MB |
| The local classifier | about 812 MB |
Without them C5 still works; those features wait for their model, and each model can be placed on the machine by hand instead.
If you run your AI models on your own computer and place the downloads above by hand, C5 needs the internet only for an occasional licence renewal — or not at all with an air-gapped licence (see C5 needs internet just to stay licensed?).
Does my data go to the cloud? #
No. Your prompts, files, and results stay on your computer, encrypted. See Privacy.
What does it cost to run? #
C5 itself is licensed software. On top of that you pay whichever SI provider you use, based on how much you use it. Running models locally costs nothing per use.
Set a budget on day one — see Budgets.
Can I use it on more than one computer? #
Yes. Install it on each and run growther activate on each. Your plan sets how many
machines you can have active.
Using it #
How do I get better results? #
Say what "done" looks like, attach the files, and say what to avoid. See Chat for the habits that help most.
Why is it asking me before doing things? #
That is the permission system protecting you. You control how much it asks — see Permissions.
Can I stop it mid-task? #
Yes. Click Stop and it stops right away. Stop is a brake, not an undo — anything already done stays done: a file that was written is still written, an email that was sent is still sent. See Chat.
Where do my finished files go? #
The Library. Everything has a Download button.
Why is a task taking so long? #
Usually it split into many smaller pieces, or it is waiting on you. Open Productivity and click the task to see exactly where it is.
How do I start C5 if it is stopped? #
You have three easy ways:
- From your browser: Click the green Start C5 button in the sidebar menu or in Settings → System.
- From your desktop: Double-click the Growther.si C5 icon in your Applications folder or Start menu.
- From your terminal: Run
growther start(orgrowther).
Models and money #
Which model should I use? #
Start with the defaults. When you want to tune it, see Choosing models.
Can I run models on my own machine? #
Yes. It costs nothing per use and nothing leaves your computer. See Model providers.
How do I keep costs down? #
Set a budget, use smaller models for simple work, be specific in your requests, and check your schedules — a frequent schedule is usually the biggest line on the bill.
The local classifier #
What is the local classifier? #
A small model, decider-0.8b, that C5 runs on your own computer to answer some of the routine questions it asks itself while it works — for example, which categories a new task belongs to. Without it, C5 sends each of those questions to the AI provider you configured. It is the Built-in model in Settings › Classifier (it used to be the Classifier card in Settings › Learning, called Local Classifier before C5 could also use a model from your providers) — see Settings › Classifier. The full guide is Classifier.
Does it change what C5 does? #
Not at first. When it is first on, it only records its answers beside the usual ones: every kind of decision it can answer starts in Shadow. C5 lets it decide something only after it has been measured, on your own install, against the answers people gave or the outcomes that followed — and only for the kinds of decision allowed under Settings › Classifier. Until then, everything is decided the way it was before. See When it starts to act.
Will it slow my computer down? #
It is built not to. On an Apple-silicon Mac it uses the GPU and answers in a fraction of a second. Where it runs on the CPU — every other machine, and a Mac whose GPU is left to a local model such as LM Studio — step quality, the kind of decision that arrives in bursts, records only one question in four, always the same ones, so for it the classifier uses about a quarter of the processor time it otherwise would. Every other kind is recorded every time. The card in Settings › Classifier says Sampling 1 in 4 on this machine under Step quality where this applies. It never holds up your work while it only records: those questions take turns in the background, one at a time, and none starts while C5 is waiting for an answer it will use. The one wait is for a kind of decision set to act (Assisted or Authoritative): if a recorded question is already being answered, C5 waits at most 1.5 seconds in all for the classifier, then carries on with the usual model's answer. It leaves two of your processor's cores alone. See On a machine where it runs on the CPU.
Is anything sent to Growther.si? #
No. The classifier runs on your machine, in its own process. Its answers, what it learns and how often it is used stay there. The one network request it involves is the model's one-time download (about 812 MB), which goes from your machine to Hugging Face, through your proxy settings if you have them. Nothing about the model, or your use of it, goes to Growther.si. See Privacy on the classifier's page.
If an administrator chooses a model from your providers for these questions, they go from your machine to that provider like any other work you send it — still not to Growther.si.
Why is it on by default? #
Because the built-in model uses no provider and no account of yours, costs nothing per use, and learns only from what it records while it is on. When it comes on, administrators get one notification saying so, and what will happen on this machine — whether the model will be downloaded, is already there, or cannot be downloaded here and why. A machine that cannot run it is told nothing, and nothing is downloaded there.
The model's publisher says its training mixture includes datasets released for research use. C5 shows that notice beside the model in Settings › Classifier. If it does not fit your use, turn the classifier off. See Why it is on by default.
How do I turn it off? #
In Settings › Classifier, switch off Built-in model. C5 asks first. Turn off leaves the model on disk, so turning it back on costs no download; then select Save to turn it off. Turn off and delete the model turns it off and frees the space at once, without Save; it is offered only to an administrator signed in at the machine itself (not from a paired phone). Anyone with permission to change settings can turn the classifier off. With it off, an administrator at the machine can still press Delete the model (… MB) on the card later. Step by step: Turn the classifier off and Delete the model.
Off means off: C5 then loads nothing for it, downloads nothing for it, and prints nothing about it. Everything else in C5 carries on as before.
To keep it off before it ever downloads — on a new install, or across a fleet — have your
organisation set the managed-policy key classifier_enabled to false. A policy can lock
it on or off; the switch then shows the lock and cannot be changed in C5, and a change to
the policy takes effect without a restart. See
Keep the local classifier on or off.
It says "Unsupported". Is something wrong? #
No. The classifier needs a supported platform, at least 8 GB of memory that the system can use, rounded to the nearest gigabyte (most 8 GB machines qualify; one that keeps more than about half a gigabyte for built-in graphics may not, and the card then shows the usable figure), and a recent enough operating system and system libraries. Where those are missing, C5 downloads nothing and works as it did before. The card gives the exact reason; see Machines it cannot run on and Common issues.
Can I use it on a machine with no internet? #
Yes. Download the model on a machine that can reach Hugging Face, copy it to the
classifier's models folder, and run growther classifier status to check it. The exact
address, size, sha256 and folder for each platform are in
Offline and air-gapped installs.
You do not need to restart C5: open Settings › Classifier, and C5 checks the new file and starts the classifier. (C5 looks for a new model file whenever that page asks for the classifier's status.) Otherwise it picks the file up at its next start.
The terminal commands are growther classifier status (can this machine run it, and is
the model there and checked) and growther classifier pull (download it now, even if
automatic downloads are switched off with GROWTHER_CLASSIFIER_NO_AUTO_DOWNLOAD=1). The
pull is refused while the egress posture is Offline, or when your organisation's policy
keeps the classifier off.
A classifier model from your providers #
Can the classifier use a model I already run? #
Yes. An administrator at the machine can choose a model that one of your providers serves —
on llama.cpp, Ollama, LM Studio or vLLM, or a cloud API such as OpenAI, OpenRouter or
DeepSeek — under Settings › Classifier › Choose a model…. C5 tests it first,
and asks it before the built-in model. While the built-in model is on, it answers whenever
the chosen model cannot, unless Use as fallback, under its switch, has been turned off;
with the built-in model off, the chosen model is the only one, and when it cannot answer C5
answers as it did before. The chosen model has its own switch, to turn it off for a while
without removing it. A new model only records its answers until it has been measured on your
install, like the built-in model was. Questions about whether a tool call is safe are asked
only of a model C5 can show runs on this machine (a llama.cpp server that hands part of its
model to other computers with --rpc can look as if it does; see the page below). See
A model from your providers.
Why can't I choose OpenClaw, PicoClaw or Hermes Agent? #
They are agent gateways, not models. A gateway hands each question to an agent, and the agent may pass it on to any model it likes. The classifier needs one model's own probabilities for each possible answer, and a gateway cannot give those. The list names them under Not listed for that reason. Choose the model the gateway uses instead, from that model's own provider — for example the same model in Ollama or LM Studio.
Why does LM Studio need "Your network"? #
Models allowed from starts at This machine, which lets C5 ask only a model it can show runs on this machine. LM Studio answers at this machine's address, but it can serve a model that actually runs on another device (LM Studio calls this LM Link), and nothing C5 can read from it says which. So C5 counts every LM Studio model as on your network. To use one, an administrator at the machine sets Models allowed from to Your network on the Classifier card, then chooses the model and confirms the screen that says where it runs and what C5 would send. If your organisation's policy locks that setting, only the policy can change it. The same goes for vLLM.
Why can't I choose a model from Anthropic, Groq, Mistral, Cohere or Amazon Bedrock? #
Their APIs do not return the probabilities the classifier reads for each possible answer, so none of their models can answer its questions. The list shows each of them with the reason. OpenAI, MS Azure, OpenRouter, DeepSeek, xAI Grok, Alibaba Cloud Qwen, Moonshot Kimi, Meta and Google Gemini can be tried, with Models allowed from set to Anywhere; they bill your account for each question, and C5 shows what a test costs before it sends anything.
Data and safety #
What if I lose my computer? #
Getting your work back depends on whether you kept a backup — see Backups and recovery.
Keeping it private is a separate question, and the honest answer is that C5's own encryption does not cover this case: your key file sits on the same drive as your data, so someone with the drive has both. Turn on full-disk encryption — FileVault on macOS, BitLocker on Windows, LUKS on Linux. That is the tool that protects a lost laptop. See Local encryption.
Can Growther.si see my work? #
No — your prompts, files, and results never leave your computer. If you connect to Mothership, what is shared is limited to measurements — the full list is at Privacy. Never connect, and nothing leaves at all.
Can I get my data out? #
Yes, at any time, from Settings. It is yours.
What happens if I stop paying? #
Your data stays on your computer and you can still export it. You own it — that does not change.
C5 itself is licensed software, so an inactive license means the app stops running. Your work is not held hostage: export what you need from the Library, or from Settings, at any time.
How do I get a license? #
Get one from growther.si. Then run growther activate on each
computer to pair it — see Connecting.
I have a referral link or discount code — how do I use it? #
Open the link and sign in when asked, or — if you only have the code — sign in to license.growther.si and enter it under Have a referral code? in the cart on the Plans page. The cart shows what comes off before you pay. Everything else — time limits, who gets what, and how to become a referrer yourself — is on the Referrals and discount codes page.
Why does C5 check in with the license service? #
C5 holds a short-lived licence file and renews it quietly in the background, well before it runs out. It is a routine check-in, not a re-purchase, and you do not have to do anything.
Renewing is also how a change reaches you: buy more seats, add a feature, or extend your term, and it appears without reinstalling anything.
C5 needs internet just to stay licensed? #
Only briefly, and only now and then. The licence file lasts long enough to cover ordinary time offline — travel, a flaky connection, a laptop shut for a week. Nothing stops the moment you disconnect.
If a machine will be offline for a long stretch, or permanently, ask us about an air-gapped licence instead.
Does my license expire soon? It shows a date only a couple of weeks away #
It should not, and if you see one, tell us.
There are two different dates involved, and only one of them concerns you: when your access ends, which is the date the app and the admin console show. The other is an internal renewal deadline for the licence file itself, a couple of weeks out and refreshed continuously. You should never be shown that one.
Still stuck? #
Try Common issues, or Getting help.
