# Growther.si > C5 runs 200+ trained SI specialists across 17 divisions on your machine — local-first, encrypted at rest, works offline. You hold the keys. Mothership adds a fleet-wide learning flywheel: proven improvements travel between deployments — scrubbed of secrets and PII, reviewed by the owner, reversible. C5 is a desktop SI harness orchestration platform for macOS, Windows, and Linux. It plans, staffs, executes in parallel, verifies, and reworks — orchestration intelligence, 35 layers deep. Bring your own model: 14+ cloud providers or fully local runtimes (Ollama, LM Studio, llama.cpp). Humans stay in command: autonomy dial, batched approvals, audit trail, provenance, killswitch. Free to download and start; licenses and pricing at https://license.growther.si. ## Key facts - **What it is**: a desktop application that orchestrates SI agents, not a chat wrapper and not a hosted service. - **Where it runs**: your own machine — macOS, Windows, Linux. The database is encrypted at rest (SQLCipher) and it works fully offline. - **Fleet**: 200+ trained specialist agents across 17 divisions, staffed automatically per task. - **Models**: bring your own — 14+ cloud providers (Anthropic, OpenAI, Gemini, Bedrock, Mistral and more) or fully local via Ollama, LM Studio, llama.cpp. Your content goes to the provider you configure, under your own account. - **Context import**: reads the Claude (when the Claude app is installed on the same computer), ChatGPT Codex and Google Antigravity sessions already on your machine, converts them to markdown, and indexes them for hybrid keyword + vector recall. Opt-in per tool, refreshed daily, nothing uploaded. - **Human control**: autonomy dial, batched approvals, provenance on every action, full audit trail, killswitch. - **Mothership (optional)**: a fleet-wide learning flywheel. Only C5's own internal instructions, once its learning loop has proven an improvement, travel between deployments — scrubbed of secrets and personal data, reviewed by the owner, reversible. Never your prompts, documents or conversations. - **Price**: licensed — perpetual or subscription plans at https://license.growther.si; free-days promotions run from time to time (see Pricing). - **Company**: CYBRNTK LLC, a California limited liability company, doing business as Growther.si (also operating as GrowtherSI and GrowtherAI). Contact help@growther.si. ## Stories - [Command your fleet — Mission Control](https://growther.si/fleet): Two hundred trained SI specialists across 17 divisions, running locally under your command. Plans, staffs, executes, verifies — with an audit trail and a killswitch. - [One prompt. An entire organization.](https://growther.si/org): C5 turns a rough prompt into a full spec, staffs the right specialists, runs the work in parallel, and hands back verified deliverables. - [Every employee, multiplied — SI at enterprise scale](https://growther.si/scale): From a ten-seat pilot to a 1,500-seat fleet on the same rails: durable execution, adaptive parallelism, SOC 2-aligned governance, and audit exports. - [While you slept, your team shipped](https://growther.si/day): Standing orders run overnight — even offline — and only wake you when a human is needed. Batched approvals, real memory, professional deliverables. - [Intelligence that compounds — the Mothership flywheel](https://growther.si/flywheel): Every C5 learns from every run. Proven improvements travel the fleet — scrubbed of secrets, reviewed by you, reversible. Compounding without surrender. ## Frequently asked questions ### Is my data private? Yes — C5 runs on your machine with its database encrypted at rest, and it works fully offline. If you enable Mothership, only proven prompt-level improvements leave — scrubbed of secrets and private data — never your documents or conversations. ### Which AI models can I use? Bring your own: 14+ cloud providers (Anthropic, OpenAI, Gemini, Bedrock, Mistral, and more) or fully local models via Ollama, LM Studio, or llama.cpp. C5 routes each task to the best-fit brain, cost-aware. ### Do I lose the work I've already done in other AI tools? No. C5 reads the Claude, ChatGPT Codex, and Google Antigravity sessions already sitting on your own machine, turns them into clean documents, and indexes them into its local memory — so months of prior thinking is searchable from day one. It refreshes daily, you turn it on per tool, and nothing is uploaded to do it. ### What happens when I'm offline? C5 keeps working. It's offline-first by design — local models keep executing, schedules keep running, and everything syncs up when you're back. ### Do agents act without my approval? You set the autonomy line. Big calls always wait for you, everything is logged with provenance, and a killswitch is always within reach. ### How do I get started? Download C5 for macOS, Windows, or Linux, pick a starting workspace, and type your first request. The core five wake up first; the rest of the fleet staffs itself. ### Can I control the self-learning? Fully. Dials for how and when it learns, plus metrics and telemetry that show exactly what the learning loops are doing in your own deployment. ### Can you help us get set up? Yes — Growther.si personalizes the installation and integration for your business or personal deployment, rapidly builds your enterprise (or life) context into C5's memory system, and offers corporate and personal training so your productivity moves at light speed. Use Get in touch → pick 'Onboarding & training.' ## Guides — local, private and scheduled AI agents ## A local AI agent that runs on your own computer Source: https://growther.si/local-ai-agent A local AI agent is software that plans and finishes multi-step work on your own computer instead of a vendor's cloud. C5 by Growther.si keeps three of an agent's four layers — app, data and memory — on your computer; the model is your choice, local or cloud. It ships as one file for macOS, Windows and Linux, runs 200+ specialist agents and, with a local model, works offline. ### What “local” actually means “Local” gets used loosely, so it pays to ask which parts of an agent actually sit on your machine. An agent has four layers, and each one either lives on your computer or somewhere else. With C5, three of the four are always local — and the fourth is your choice. _The app, its data and its memory always run on your computer. The model is local with Ollama, LM Studio or llama.cpp, or a cloud provider you pick._ | Layer | With a local model | With a cloud model | | --- | --- | --- | | The agent app (planning, tools, schedules) | Your computer | Your computer | | Files, tasks and history | Your computer, encrypted | Your computer, encrypted | | Memory and search index | Your computer | Your computer | | The model doing the thinking | Your computer | The provider you chose, on your own account | That last row is the one to watch: the moment a task uses a cloud model, the text sent for that task reaches that provider. Run a [local model](https://growther.si/local-llm-agent) and nothing has to leave the machine — see [what a private AI agent means](https://growther.si/private-ai-agent). ### Can an AI agent run locally on my computer? Yes, on an ordinary laptop or desktop. C5 ships as one file for macOS, Windows and Linux — no Node, no Docker, nothing else to install first. It runs its agents, its schedules and its database on the machine it is installed on, and it keeps going [without an internet connection](https://growther.si/offline-ai-agent) when it has a local model to think with. - **macOS, Windows and Linux** — one installer, or Homebrew on macOS and Linux. - **No GPU required for C5 itself.** The app is light; a local model's speed depends on your hardware, and cloud models need none. - **Mix and match.** Route private work to a local model and heavy lifting to a cloud model, task by task. ### What a local AI agent can do A chatbot answers. An agent finishes. Give C5 a goal in plain words and it writes the spec, picks the right specialists from 17 divisions, runs them in parallel, checks the result and reworks anything that misses — see [multi-agent orchestration](https://growther.si/multi-agent-orchestration). - **Finished deliverables** — reports, plans, decks and publication-grade PDFs, not drafts. - **Recurring work** — [scheduled jobs](https://growther.si/scheduled-ai-agent) that run daily, weekly or monthly on their own. - **Work across your computer** — tools, integrations, MCP servers and, when nothing else will do, [computer use](https://growther.si/ai-automation-on-your-computer). - **Your own files** — an encrypted, searchable memory of [your local data](https://growther.si/ai-agent-local-data). ### Local vs cloud AI agents | | C5 (local) | Typical cloud AI agent | | --- | --- | --- | | Where tasks run | Your computer | The vendor's servers | | Where files and history live | Your disk, encrypted at rest | The vendor's storage | | Works offline | Yes, with a local model | No | | Model choice | 14+ cloud providers or fully local | The vendor's models | | Who holds the keys | You | The vendor | A local agent is the wrong fit if no computer of yours can stay on for the work — schedules wait while the machine sleeps — or if you want a hosted service someone else keeps running. Already using a cloud agent? See [a private alternative to cloud AI agents](https://growther.si/cloud-ai-agent-alternative) — you can even [bring your Claude, Codex and Antigravity sessions](https://growther.si/import-claude-codex-history) with you. ### Get started 1. **Download C5** — Download C5 for macOS, Windows or Linux and activate it with your licence. 2. **Connect a model** — Point C5 at a local runtime such as Ollama, LM Studio or llama.cpp, or add your own key for a cloud provider. 3. **Pick a starting workspace** — Choose the workspace that matches the work — each one carries its own rules of engagement. 4. **Type your first request** — State the outcome you want. C5 turns it into a spec, staffs the specialists, runs them and verifies the result before handing it back. ### Frequently asked questions **What does a local AI agent cost to run?** It depends on the models you use. C5 itself needs a licence — plans are on the [Pricing](https://growther.si/#modal=pricing) tab. Local models add no per-use cost, because they run on your own hardware; cloud providers, if you use them, bill you directly on your own account. **Do I need a GPU to run a local AI agent?** No, not for C5 itself. A GPU or Apple silicon makes local models faster, and smaller local models run on ordinary laptops. If your machine is modest, use a cloud model for heavy tasks and a local one for private ones. **Which local models work with C5?** Anything you can serve through Ollama, LM Studio or llama.cpp, plus local agent gateways such as OpenClaw, PicoClaw and Hermes Agent. C5 routes each task to the best-fit model you have allowed. **Does a local AI agent work without internet?** Yes, with a local model. C5 is offline-first: agents keep executing and schedules keep running with no connection, and anything that needs the network catches up when you reconnect. **Can a local AI agent still use ChatGPT or Claude models?** Yes. Add your own key for OpenAI, Anthropic, Gemini, Bedrock, Mistral or another of the 14+ supported providers. The agent stays local; only the text sent for that task goes to that provider. --- ## A private AI agent: your data stays on your machine Source: https://growther.si/private-ai-agent A private AI agent does its work without storing your files, history or memory on a vendor's servers. C5 by Growther.si keeps all of it on your computer, encrypted at rest, and lists exactly what crosses the edge of the machine. With a local model through Ollama, LM Studio or llama.cpp nothing has to leave; with a cloud model, only the text that task sends reaches that provider. ### What stays on your machine, always The things you would least want on a vendor's server never go to Growther.si, whichever model you use. C5 is a [local AI agent](https://growther.si/local-ai-agent): the app, its databases and its memory run on the computer it is installed on. - **Your prompts and chats** — what you ask for, and every conversation and its history. - **Your files** — anything you attach, and everything your agents create. - **Your results** — deliverables, documents, notes and raw task output. - **Your API keys** — stored encrypted on your computer and only ever sent to the provider they belong to. - **Your context** — the standing instructions you wrote, and the search index of your work. They live in four encrypted databases in your data folder, by default ~/.growther. Growther.si cannot read them, because it never receives them. ### What leaves, and when A private AI agent should be able to tell you exactly what crosses the edge of your machine. This is the whole list for C5. | Situation | What leaves | Where it goes | | --- | --- | --- | | A task on a local model | Nothing about the task | Stays on your computer | | A task on a cloud model | The text sent for that task | That provider, on your own account and under its terms | | A tool an agent uses (web search, an API) | The request, with credentials and personal data redacted by default | That tool's endpoint | | Sharing improvements with Mothership (optional) | Measurements of how well work went, and C5's own internal instructions once its learning loop has improved them, scrubbed of secrets and private data | Growther.si — never your prompts, documents or conversations | | Remote Control through an issued address (off by default) | Your session, encrypted to and from Cloudflare | Cloudflare's network, on the way to your own machine | The second row is the honest boundary. Encryption at rest protects data sitting on your disk; it cannot change what a cloud provider receives once a task sends it. Fully private means a local model — see the [local LLM agent](https://growther.si/local-llm-agent) page for Ollama, LM Studio and llama.cpp. ### Private by design, not by policy A privacy policy is a promise about data a vendor already holds. C5's approach is that the vendor never holds it, and the controls sit with you. - **Encrypted at rest.** Chats, tasks, files, keys and settings are encrypted in local databases. There is no master key and no back door. - **Your own keys.** Cloud providers bill your account directly, and requests go from your machine to the provider, not through Growther.si. - **Checks before anything goes out.** Before a tool runs, C5 scans what it is about to send for API keys, tokens, card numbers, IBANs and other personal data, and redacts them by default. The checking happens on your machine. - **A killswitch and an audit trail.** One control stops every agent at once, and every action is logged with provenance. - **Your data stays yours.** Export, decrypt or delete it at any time, from Settings. One limit worth knowing: the key that unlocks the databases is a file on the same drive. For a lost or stolen laptop, turn on full-disk encryption — FileVault, BitLocker or LUKS. The [encryption docs](https://docs.growther.si/c5/security/encryption) explain exactly what C5's own encryption does and does not cover. ### A private AI agent vs a cloud agent's privacy settings | | C5 with a local model | C5 with a cloud model | Typical cloud AI agent | | --- | --- | --- | --- | | Files, history and memory | Your disk, encrypted | Your disk, encrypted | The vendor's storage | | The text a task sends | Stays on your computer | The provider you chose | The vendor | | Works without internet | Yes | No, for that task | No | | Who holds the keys | You | You | The vendor | C5 is the wrong fit if you have no hardware that can run a local model and a rule that nothing may reach any cloud provider: the fully private setup needs a local model to think with. Moving from a hosted agent? The [private alternative to cloud AI agents](https://growther.si/cloud-ai-agent-alternative) page covers the switch, including bringing your existing history with you. ### Set up a fully private AI agent 1. **Install C5** — Install C5 on macOS, Windows or Linux with one command, and activate it once while you are online. 2. **Connect a local model** — In Settings, open Integrations, choose the local option and point C5 at Ollama, LM Studio or llama.cpp. C5 checks the connection and lists the models it finds. 3. **Make it the only provider** — Leave cloud provider keys out, so every task runs on the local model. 4. **Turn on full-disk encryption** — Use FileVault, BitLocker or LUKS so a lost laptop does not expose the data folder and its key together. 5. **Optionally switch C5 offline** — An administrator can set the egress posture to Offline, which stops every connection C5 makes on its own, and the agent keeps working [without internet](https://growther.si/offline-ai-agent). ### Frequently asked questions **Is C5 really private if I use OpenAI or Claude?** It depends on what you count. Your files, history, memory and keys stay on your computer, encrypted; the text a task sends to a cloud model reaches that provider, on your own account and under its terms. For work that must not leave the machine, route it to a local model. **Does Growther.si see my data?** No. Your prompts, files and results never reach Growther.si. If you choose to share improvements with Mothership, what is shared is limited to measurements of how well work went and improved versions of C5's own internal instructions — never your prompts, documents or conversations. **What does Mothership share?** Only measurements and C5's own instructions, and only if you opt in. It covers your version, your licence state, how often work succeeded and how long it took, and C5's internal instructions once its learning loop has measurably improved them, scrubbed of secrets and private data. If a summarizing task ran, it may share that it succeeded in 40 seconds — not the document or the summary. **Can I use a private AI agent for confidential client work?** Yes, with a local model: the client material, the work and the deliverables stay on your machine. With a cloud model, check that the provider's terms fit your client agreement. Deliverables can also be signed, so a client can check a file came from you unchanged. **Where is the data stored?** On your own disk, in your data folder — by default ~/.growther on macOS and Linux and under your user profile on Windows. Databases are encrypted and are only allowed on a local disk; C5 refuses network folders and synced folders such as OneDrive, Dropbox or iCloud Drive. --- ## An AI agent that keeps working offline Source: https://growther.si/offline-ai-agent An offline AI agent keeps planning and doing multi-step work when there is no internet connection. C5 by Growther.si is offline-first: with a local model through Ollama, LM Studio or llama.cpp, its agents keep executing and its schedules keep firing on your computer. Anything that needs the network — a cloud model, web search, a licence refresh — waits and catches up when you reconnect. ### What works offline Losing the connection does not stop the parts that already live on your machine. C5 is a [local AI agent](https://growther.si/local-ai-agent), so its database, its schedules and its search index are there whether or not the network is. With a local model to think with, all of this keeps going: - **Tasks and agents** — planning, staffing, running and checking work, on the local model. - **Schedules** — standing orders still fire on time, overnight included. - **Your files and memory** — the Library, attachments and keyword search; semantic search too, once its models are on the machine. - **Voice** — offline speech to text, once its small model is downloaded. - **Safety checks** — redaction of credentials and personal data runs on your machine, so it works the same offline. ### What still needs a connection Offline-first does not mean nothing ever touches the network. This is the full list, and what happens to each while you are offline. | Needs a connection | Why | While offline | | --- | --- | --- | | Cloud models | The provider runs the model | Tasks routed to a cloud model wait; route them to a local model instead | | Web search and online tools | They reach an outside endpoint | Those tool calls cannot complete | | Installing, activating and updating | Downloads and the one-time licence pairing | Do these while online | | Licence renewal | C5 renews a short-lived licence file in the background | The licence covers ordinary time offline, such as travel or a week away | | Memory-search, voice and classifier models | Downloaded once from Hugging Face | Already-downloaded models keep working; missing ones can be placed by hand | | Remote Control through an issued address | It runs through Cloudflare's network | Use your own network or SSH instead | ### Set up an AI agent for offline use 1. **Install and activate while online** — Install C5 on macOS, Windows or Linux and run its one-time activation, which needs the licence service. 2. **Connect a local model** — In Settings, open Integrations, choose the local option and point C5 at Ollama, LM Studio or llama.cpp. C5 checks the connection and lists the models it finds. 3. **Let the one-time downloads finish** — C5 fetches its memory-search models (about 2.26 GB) and the offline speech model (about 60 MB) in the background after the first start. Keyword search works before they arrive. 4. **Make the local model the first choice** — Put the local model first in Coordinator Priority under Settings › Integrations, so tasks do not wait on a cloud provider when the connection drops. For the model choices and what each needs from your hardware, see the [local LLM agent](https://growther.si/local-llm-agent) page. ### Travel, field work and air-gapped sites A laptop on a plane, a site with poor coverage and a network that is never connected are three different cases, and C5 handles each. - **Travel and patchy connections.** Nothing stops the moment you disconnect. Work routed to a local model carries on, and anything that needs the network catches up later. - **Fully offline on purpose.** An administrator can set the egress posture to **Offline**, which stops every connection C5 starts on its own — licence refresh, update checks, model downloads and platform check-ins. It can be locked by policy across a fleet. - **Air-gapped machines.** Place the models by hand from a connected machine — each has a published address, size and sha256 — and ask Growther.si for an air-gapped licence for a machine that will stay offline. Going offline is also the simplest way to make a [private AI agent](https://growther.si/private-ai-agent) provably quiet. For a server or a locked-down site, the [self-hosted AI agent](https://growther.si/self-hosted-ai-agent) page covers installs, policy and remote access. ### Frequently asked questions **Can an AI agent work without internet?** Yes, if both the agent and its model run on your computer. C5 runs locally and works with local models through Ollama, LM Studio or llama.cpp, so its agents keep executing and its schedules keep running with no connection. **Do scheduled tasks still run when I am offline?** Yes. Schedules run on your machine, so they fire on time without a connection. A scheduled task that is set to use a cloud model needs the network for that step; give it a local model if it must run offline. **How long can C5 stay offline?** Long enough for ordinary time away — travel, a flaky connection, a laptop shut for a week. C5 renews a short-lived licence file quietly in the background while it has a connection. For a machine that will be offline for a long stretch or permanently, ask Growther.si for an air-gapped licence. **Can I run C5 on an air-gapped network?** Yes, and it is a supported way to use it. Set the egress posture to Offline, place the memory-search, voice and classifier models by hand from a connected machine, and use an air-gapped licence. LDAP sign-in to a directory on your own network still works. **What happens to a cloud-model task when the connection drops?** It waits. That step cannot reach the provider until you reconnect. If you have a local model connected and placed in the coordinator order, C5 can fall back to it rather than stopping. **Does offline work continue if my laptop is closed?** No. Offline means no network, not no computer: C5 runs on the machine itself, so if it is asleep the schedule waits and catches up when it wakes. For work that must run overnight, keep the machine awake or use one that stays on — the [scheduled AI agent](https://growther.si/scheduled-ai-agent) page covers the trade. --- ## A self-hosted AI agent on hardware you control Source: https://growther.si/self-hosted-ai-agent A self-hosted AI agent runs on hardware you own or control instead of as a vendor's hosted service. C5 by Growther.si installs with one command on macOS, Windows or Linux, needs no Docker, and listens on loopback until you decide who else may reach it. Reach it from other devices over your own network, over SSH with growther connect, or through Remote Control. ### Install a self-hosted AI agent with one command You need nothing installed first — not Node, not Docker. C5 is a single program file that brings everything it needs. On macOS and Linux run **curl -fsSL https://growther.si/install.sh | bash**; on Windows, in PowerShell, run **irm https://growther.si/install.ps1 | iex**; or use Homebrew with **brew install GrowtherSI/tap/growther-c5**. The installer picks the right build for your machine, checks the download against its SHA-256 fingerprint and stops if a single byte is wrong, puts the growther command on your PATH, and creates the data folder. | System | Chips | Oldest version supported | | --- | --- | --- | | macOS | Apple silicon and Intel | macOS 13.5 Ventura | | Windows | x86_64 and ARM64 | Windows 10 (Windows Server 2016) | | Linux | x86_64 | glibc 2.28 — Ubuntu 20.04, Debian 10, RHEL 8 and newer | 1. **Run the installer** — Run the one-line installer for your system, or install through Homebrew. 2. **Check the install** — Run growther --help. If you see the list of commands, C5 is installed. 3. **Activate this machine** — Run growther activate. C5 shows a short code and opens a page where you confirm it. This happens once per computer. 4. **Start C5** — Run growther. C5 starts and opens the app at http://localhost:4299. Run growther service install if you want it to start by itself. ### Run it on a server or a headless box The first thing to settle on a server is who can reach the install. By default C5 listens on loopback only, so it is reachable from the machine it runs on and nowhere else — the right default for a laptop, and a safe starting point for a server. - **No desktop needed.** On a Linux server without an applications menu, start C5 with the command. growther login signs in from the command line on a machine with no browser. - **Keep it running.** growther service install starts C5 automatically; growther status and growther stop do what they say. - **Serve it to your network deliberately.** Set GROWTHER_BIND_HOST, list the names it may answer to in GROWTHER_ALLOWED_HOSTS, and put it behind a reverse proxy that terminates TLS. - **Data stays on local disk.** Databases are refused on network shares, mapped drives and synced folders, and two machines never share one data folder. Network storage is welcome for encrypted backups. For the full set of bind, host and origin settings, see [Network and egress](https://docs.growther.si/c5/configuration/network) in the docs. ### Three ways to reach your install | Way in | How you get there | Who is in the path | | --- | --- | --- | | Your own network, or a name and certificate you supply | Straight to the machine | Nobody but you and the machine | | Your own SSH connection, with growther connect | A local port forwarded over SSH you already have | Nobody but you and the machine | | An address Growther.si issues | A tunnel Growther.si creates for your install | Cloudflare, with the connection encrypted to and from it | The first two need nothing from Growther.si and keep working without it. The third works from anywhere without touching your router; Remote Control is off until you turn it on at the machine itself. See [AI agent remote control](https://growther.si/ai-agent-remote-control) for pairing a phone and managing devices. ### Managed configuration for IT For a fleet, C5 ships a Windows Installer package (.msi) and a macOS installer package (.pkg) that wrap the same signed program, and it takes its settings from the tools IT already runs. - **Policy from Intune, Group Policy, Jamf or a signed policy file.** Locked settings show as managed by your organisation on every user's screen. C5 writes the ADMX, mobileconfig and schema templates itself. - **Sign-in through your directory.** Microsoft Entra ID, Okta, Google Workspace and other OpenID Connect providers, SAML 2.0, or on-prem Active Directory over LDAP. - **Audit shipping.** Send the tamper-evident audit trail to Microsoft Sentinel, Splunk, OpenTelemetry or syslog, and export signed evidence bundles. - **What may leave, decided centrally.** Lock the network posture offline, restrict which providers and models agents may use, set a global cost cap, and control sharing with Mothership. ### Self-hosted vs SaaS AI agents | | C5 (self-hosted) | Typical SaaS AI agent | | --- | --- | --- | | Where it runs | Your machine or server | The vendor's servers | | Where data lives | Your local disk, encrypted at rest | The vendor's storage | | Who can reach it | Loopback by default; you choose the paths | Through the vendor's website or app | | Updates | growther update, checked against a signed manifest; rollback available | Whenever the vendor ships | | Works offline | Yes, with a local model | No | Self-hosting is the wrong fit if you want a service someone else keeps running, or if your users work on a shared multi-session host such as RDS, Citrix or Azure Virtual Desktop, where C5 refuses to start. Self-hosting is one part of running a [local AI agent](https://growther.si/local-ai-agent). For a machine that must make no outbound connections at all, see the [offline AI agent](https://growther.si/offline-ai-agent) page, and for keeping task data away from cloud providers, the [private AI agent](https://growther.si/private-ai-agent) page. ### Frequently asked questions **Do I need Docker to self-host C5?** No. C5 is a single program file and brings everything it needs. You do not need Docker, Node or any other runtime installed first. **Can I run a self-hosted AI agent on a VPS?** Yes, on Linux x86_64 with glibc 2.28 or newer, such as Ubuntu 20.04, Debian 10 or RHEL 8. C5 listens on loopback by default, so the simplest way in is growther connect over your existing SSH connection. ARM Linux is not supported. **Can C5 run air-gapped?** Yes. Set the egress posture to Offline, which can be locked by policy, place the memory-search, voice and classifier models by hand, and ask Growther.si for an air-gapped licence. Pair it with a local model so tasks never need the internet. **How do updates work?** With one command. growther update downloads the latest release, checks that it is genuine and installs it, restarting C5 if it was running. growther rollback goes back to the version you had. Packages deployed by IT default to not updating themselves, so the package manager stays in control. **How many users can use one install?** It depends on your plan, which sets how many machines can be active. An install can sign people in through Entra ID, Okta, SAML or LDAP, with roles and permissions. C5 refuses to start on shared multi-session hosts such as RDS, Citrix or Azure Virtual Desktop. --- ## An AI agent that works with your local data Source: https://growther.si/ai-agent-local-data An AI agent with local data reads and searches files on your own computer instead of a copy you uploaded somewhere. C5 by Growther.si keeps your attachments, notes and agent output encrypted on your disk and indexes it for keyword and semantic search on the same machine. Any agent can find what was written weeks ago, and with a local model your files never have to leave the computer. ### What the agent can read Your files do not have to be uploaded anywhere to be useful to an agent. Everything C5 works with sits in your data folder, on your disk, and agents can draw on all of it without you pointing at a file. - **Files you attach** — documents, spreadsheets, PDFs, images, code and plain text, dropped into a chat or a task. - **Folders you choose** — under Settings › Locations, tell C5 which folders to bring documents in from and where finished work should land. - **Everything agents write** — deliverables, documents, notes and raw output, kept in the Library and indexed for search. - **Your earlier AI sessions** — the Claude, ChatGPT Codex and Google Antigravity sessions already on your machine can join the same searchable memory, so what you worked out in another tool is findable here. See [importing Claude, Codex and Antigravity sessions](https://growther.si/import-claude-codex-history). ### How local search works Finding a note that contains an exact phrase and finding a note you only half remember are different jobs. Memory search does both, and both run on your machine: - **Keyword search** finds documents containing the words you typed. It works the moment C5 starts. - **Semantic search** finds documents that mean the same thing in different words — asking about “customer churn” can surface a note titled “why people cancel”. Semantic search needs three small models: one to build the index, one to put the most relevant results first, and one to expand a question into several searches. C5 downloads them once, about 2.26 GB in total, from Hugging Face, pinned to a fixed version and checked by size and sha256 before use. Nothing about that download goes to Growther.si, and an offline machine can have them placed by hand — the details are in the [memory search docs](https://docs.growther.si/c5/tools/memory-search). ### See your data Search results are only half the picture. C5's 2D/3D vector database viewer lets you visualize the data from your tasks and imported context. The Ops film above shows it, alongside budgets and audit logs. For finished work, the Library keeps Projects, Deliverables, Documents, Notes, Attachments and Output in one place, with search, date and tag filters. Its search looks inside your files, not only at their names; Attachments searches names, descriptions and C5's analysis of each file. ### Keeping your data local _Files, memory and the search index always stay on your computer. Only the model layer can be a cloud provider, and only if you choose one._ | | With a local model | With a cloud model | | --- | --- | --- | | Files, notes and attachments | Your disk, encrypted | Your disk, encrypted | | Search index and search models | Your computer | Your computer | | Passages an agent sends to the model | Stay on your computer | Reach the provider you chose, for that task | Indexing and search never leave the machine. A cloud model sees only the passages a task sends it — nothing more, but nothing less — so for sensitive material, use a local model. The [private AI agent](https://growther.si/private-ai-agent) page sets out the whole boundary, and C5 as a [local AI agent](https://growther.si/local-ai-agent) explains which layers are always local. If you want your documents searchable from a hosted service with no computer of your own running, C5 is the wrong fit: it works on the files on this machine. ### Export it, or delete it - **Download anything** — an attachment from its preview, several at once, or a whole project as a zip. - **Export everything** — decrypt and export your data at any time from Settings. - **Delete what you do not need** — deleting from the Library cannot be undone, and destroying the device key makes any surviving copy of the databases unreadable. Researchers who live in their reading notes may want the [AI agent for researchers](https://growther.si/ai-agent-for-researchers) page next. ### Frequently asked questions **Does C5 upload my files?** No. Your files and notes stay on your computer, encrypted at rest, and the search index is built and searched on your machine. If a task uses a cloud model, the passages that task sends reach that provider; with a local model, nothing has to leave the machine. **What file types can the agent read?** Most common types: documents, spreadsheets, PDFs, images, code and plain text. Attach them to a chat or task, or point C5 at the folders you work in under Settings › Locations. **Does local semantic search need the internet?** Only once, to download its three search models (about 2.26 GB) from Hugging Face. Keyword search works without them. On a machine with no internet, the models can be copied over and placed by hand. **Can it search my old Claude and Codex sessions?** Yes. C5 reads the Claude, ChatGPT Codex and Google Antigravity sessions already on your machine and brings them into its local, searchable memory, one tool at a time, with nothing uploaded to do it. For Claude, the Claude app needs to be installed on the same computer; chats on chatgpt.com are not read. **Can I delete what the agent remembers?** Yes. Delete items from the Library — this cannot be undone — or export and delete your data from Settings. Destroying the device key makes any remaining copy of the encrypted databases unreadable. --- ## Turn your local LLM into a team of agents Source: https://growther.si/local-llm-agent A local LLM agent puts a model running on your own hardware to work on multi-step tasks, not just chat. C5 by Growther.si connects to Ollama, LM Studio and llama.cpp and gives that model a job: a spec, a team of specialists, a check on the result. It routes each task to the best-fit model by skill, speed and cost, mixing local and cloud models only where you allow it. ### From a chat window to finished work A runtime such as Ollama gives you a model and a prompt box. C5 gives that model a job. Describe the outcome in plain words and C5 writes the spec, staffs specialists from its 200+ agents across 17 divisions, runs them in parallel, checks the result and reworks what misses — all orchestrated on your machine. - **Deliverables, not replies** — reports, plans and publication-grade PDFs, filed in the Library. - **Schedules** — standing orders that run daily, weekly or monthly on the model you choose. - **Memory** — agents search everything earlier tasks wrote, so the local model starts with context. - **Tools** — integrations and MCP servers, behind permissions and guardrails you set. ### Local runtimes C5 works with | Runtime | What it is | How C5 uses it | | --- | --- | --- | | Ollama | A local model server | Any model you serve, as a provider | | LM Studio | A desktop app that serves local models | Any model you load, as a provider | | llama.cpp | A lightweight inference server | Any model you serve, as a provider | | Hugging Face Inferencer | Local or hosted inference | As a provider, local or cloud | | OpenClaw, PicoClaw, Hermes Agent | Local agent gateways | As a provider that hands work to its own agent | Coming from a gateway? The [OpenClaw alternative](https://growther.si/openclaw-alternative) page compares the two approaches. ### Connect a local LLM to C5 1. **Start your runtime** — Start Ollama, LM Studio or a llama.cpp server and load the model you want to use. 2. **Add it in Integrations** — In C5, open Settings, then the Integrations tab, choose the local option and point C5 at the address where your model is running. 3. **Let C5 check it** — C5 tests the connection and lists the models it finds. 4. **Set the order** — Under Coordinator Priority, choose which models C5 tries first for chats, tasks and background work. ### Mix local and cloud models You do not have to choose one model for everything. A common setup is a local model for routine and private work and a stronger cloud model, on your own key, for the hardest parts. | Kind of work | A good fit | | --- | --- | | Sorting, labeling, simple formatting | A small, fast model | | Everyday writing and summarizing | A mid-range model | | Hard reasoning, tricky analysis, planning | Your strongest model | | Bulk repetitive work | A local model, if you can | - **Fallbacks.** If a provider is down, C5 moves to the next model in your order rather than stopping, and a model that keeps failing is tried last until it recovers. - **14+ cloud providers** — Anthropic, OpenAI, Gemini, Bedrock, Mistral and more — sit alongside your local runtimes. - **Where the text goes.** On a local model, a task's text stays on your machine; route a task to a cloud model and that provider receives what the task sends it. See [what a private AI agent means](https://growther.si/private-ai-agent). ### What to expect from a local LLM - **No cost per use.** Once it is set up, a local model costs nothing beyond your hardware. - **Works offline.** With a local model, agents keep executing and schedules keep running — see the [offline AI agent](https://growther.si/offline-ai-agent) page. - **Usually less capable** than the best paid models, which is why mixing them is worth it. - **Hardware matters.** C5 itself is light; a local model's speed depends on your machine. A local-only setup is the wrong fit if every task needs your strongest reasoning and no cloud model is allowed: small models do the routine work well, not the hardest. Running your model on your own hardware is the last layer of a fully [local AI agent](https://growther.si/local-ai-agent): the app, its data and its memory are already on your computer. ### Frequently asked questions **Which local LLMs work with C5?** Any model you can serve through Ollama, LM Studio or llama.cpp, plus Hugging Face Inferencer and local agent gateways such as OpenClaw, PicoClaw and Hermes Agent. C5 lists the models it finds when you connect the runtime. **Is a local LLM good enough for agent work?** It depends on the work. For routine and repetitive tasks, often yes. Local models are usually less capable than the best paid ones, so many people route hard reasoning and planning to a cloud model and keep everything else local. **Do I need a GPU to run a local LLM agent?** No, not for C5 itself. A GPU or Apple silicon makes local models faster, and smaller models run on ordinary laptops. If your machine is modest, use a cloud model for heavy tasks and a local one for private ones. **Can I use Ollama and a cloud model together?** Yes. Connect both, set the order under Coordinator Priority, and C5 routes each task to the best-fit model you have allowed. If one provider is down, it falls back to the next. **Does anything leave my machine with a local LLM?** Nothing about the task. C5 itself still makes a few connections of its own — licence renewal, update checks and one-time model downloads — and an administrator can stop all of them with the Offline egress posture. --- ## A scheduled AI agent that does recurring work on its own Source: https://growther.si/scheduled-ai-agent A scheduled AI agent runs a task at a time you set — once, or every day, week, month or year — without being asked. C5 takes the schedule in plain words, as you would ask in chat, and runs it on your own computer, offline with a local model. Anything that needs a decision waits for you in one review window. ### Good things to schedule If you ask for the same thing over and over, it should be a schedule. Set it up once and C5 does it again at the time you chose, with the same specialists and checks as any other task. - **A weekly summary** every Friday afternoon. - **A morning check** of your inbox or a news topic before you start work. - **A month-end report** pulled together on the first of the month. - **A nightly tidy-up** of a downloads or project folder. - **ROI reports** — C5's built-in trackers audit your own tasks, deliverables and costs weekly, monthly, quarterly and yearly. ### How to set up a scheduled AI agent 1. **Open Schedules** — Go to Schedules in the C5 sidebar and click New Schedule. 2. **Choose what it does** — Pick Create Task to run a full task, Use Recipe to follow a recipe C5 saved from earlier work, or Send Request to get a single answer back as a notification. 3. **Describe the work** — Write what you want done in plain words, the same way you would in Chat. You can also dictate it. 4. **Pick when it runs** — Choose a date, or a simple repeat such as every day, every Monday or the first of the month. A cron pattern is there if you need exact timing, and C5 shows the next few run times before you save. 5. **Save** — Save the schedule. Each run appears on the task board like any other task, and its history shows when it ran, whether it worked and what it produced. ### What happens while you are not watching You will not be at the keyboard when most scheduled runs happen. Each run plans, staffs and checks its work like any task; what it may do alone, and what waits for you, is your call. - **Decisions wait in one place.** Anything that needs a person lands in the HITL review window, and several waiting tasks can be approved together. - **Auto Review for unattended work.** With it on, C5's own review approves work that passes and any follow-up recipes start straight away. With it off, each step waits for your sign-off. - **Failures do not stop the schedule.** A failed run tries again at its next normal time, and a Send Request schedule retries within minutes, up to five attempts. - **Follow-up recipes.** A schedule can line up recipes to run after each task is approved — one after another or all at once. - **Offline is fine.** With a local model, schedules keep running without a connection, and anything that needs the network catches up when you reconnect. Schedules are one part of [AI automation on your own computer](https://growther.si/ai-automation-on-your-computer). To check on a run from your phone, use [remote control for your AI agent](https://growther.si/ai-agent-remote-control). ### Scheduled AI agent vs cron and cloud automation | | C5 schedule | Cron job | Typical cloud automation service | | --- | --- | --- | --- | | How you describe the job | Plain words, like a chat request | A command or script you write | Trigger and action steps you configure | | What runs | A team of agents that plans, works and checks the result | Exactly the command you wrote | The steps you built, in order | | Where it runs | Your computer | Your computer or server | The vendor's servers | | Your computer must be awake | Yes — it waits and catches up when it wakes | Yes | No | | Asks before a decision | Yes, through human review | No | Only where you build it in | The trade is honest: a cloud service can run while your laptop is closed, and C5 cannot. In return the work, its files and its history stay on [a local AI agent](https://growther.si/local-ai-agent) you control. For a business with routine reporting, see [an AI agent for small business](https://growther.si/ai-agent-for-small-business). ### Frequently asked questions **Does my computer have to be on for a scheduled AI agent to run?** Yes. C5 runs on your machine, so if it is asleep the schedule waits, and C5 catches up when it wakes. A desktop that stays on, or a laptop with sleep turned off while plugged in, suits busy schedules. **Can a scheduled AI agent run offline?** Yes, with a local model such as one served by Ollama, LM Studio or llama.cpp. C5 is offline-first: schedules keep running, and anything that needs the network catches up when you reconnect. A schedule that uses a cloud model needs a connection to reach that provider, and C5 needs the network now and then to renew its licence. **How do I see what a scheduled task did?** In its history. Each schedule records when it last ran, whether it worked, how long it took and what it produced. Click any run to open it like a normal task and see every step. Finished files are in your Library. **Can a schedule send me the result?** Yes. A Send Request schedule delivers its answer as a notification in C5. A Create Task or Use Recipe schedule puts its work on the task board and its files in your Library, and can start follow-up recipes once the work is approved. **What does a scheduled AI agent cost to run?** It depends on the model. Runs on a local model cost nothing per use; runs on a cloud model are billed by that provider on your own account, and a frequent schedule adds up — an hourly one runs about 720 times a month. Set a budget before you build a busy schedule. C5 itself needs a licence; plans are on [Pricing](https://growther.si/#modal=pricing). --- ## AI automation that runs on your own computer Source: https://growther.si/ai-automation-on-your-computer AI automation on your computer means agents do multi-step work, from files and reports to clicking through programs, on your machine, not in the cloud. C5 reaches programs through tools and integrations first and drives the screen only for programs with no other way in, recording every action. You set how far it may act before it asks you. ### Tools first, screen last The work you want done lives in files, websites and programs, not in a chat window. C5's agents read and write files, search the web, run commands and work with the programs you connect. Every tool call shows in the task, so you can see what was used, what it asked for and what came back. | Way in | What it is for | Where you set it | | --- | --- | --- | | Built-in tools | Files, web search and reading, commands, images and short videos | Settings › Extensions | | Integrations | Services you already use, connected with a key or a sign-in | Settings › Integrations | | MCP servers | New abilities through the open Model Context Protocol, switched on one by one | Settings › Integrations | | Computer use | Seeing the screen and driving the mouse and keyboard, for programs with no other way in | Settings › Tools (off by default) | Computer use is the fallback, not the default. Reach for it for an old desktop program with no way to connect to it, a website that only works by clicking through it, or copying between two programs that do not talk to each other. If a tool or integration can do the job, it is faster and safer. ### Work with the agents, or let them run Some jobs you want to watch; others you want to come back to finished. Every job lands on the Productivity board and moves from To do to Done, where you can drag cards, pause, retry or cancel, or turn on Auto Review and let a task finish on its own. One request can become many small tasks that run at the same time — that is [multi-agent orchestration](https://growther.si/multi-agent-orchestration) at work. Work you repeat belongs on a timer: [schedule an AI agent to run every morning](https://growther.si/scheduled-ai-agent) and it shows up on the same board. ### Stay in control of what it does - **A permission ladder.** Reading is allowed; writing, running commands, sending, deleting and spending ask you first by default; screen control is off. - **Your answer, your scope.** Approve once, deny, or always allow one agent one tool — for good, for an hour or for ten uses. - **Autonomy by type of work.** Set the minimum autonomy level each kind of work needs, and whether a task below it is downgraded or blocked. - **Guardrails that win.** Redaction of personal data in tool arguments and limits on where web tools may connect run whatever else is set. - **Stop and pause.** Stop halts an agent's next step at once; Master Pause halts all new task steps for everyone. - **A record of everything.** Tool decisions and permission changes are listed in Ops › Audit. All of it runs on [a local AI agent](https://growther.si/local-ai-agent): the files it touches and the screenshots it takes stay in your data folder. Route a task to a cloud model and that provider receives the text the task sends it; keep it on a local model and nothing leaves the machine — see [what a private AI agent means](https://growther.si/private-ai-agent). If you want an automation that nobody at the machine has to approve, C5 is the wrong fit: computer use stays off until an administrator turns it on there, and sending, deleting and spending ask first by default. ### Computer use, with the brakes on Some programs have no connection, no export and no API, and for those computer use is the way in. When it is on, the agent takes a picture of the screen, decides what to do, moves the mouse or types, then looks again. You see every action as it happens; nothing runs invisibly. - **Off until an administrator turns it on** at the machine, under Settings › Tools › System Control. Your operating system asks for permission too. - **Only programs you allowed.** An agent can work only with the programs you approved. - **Risky clicks ask where C5 can tell.** Actions that look like sending, deleting, publishing or paying stop for your approval first. - **Unattended runs need a limit.** A background task must carry a time or step limit before it may drive the screen. - **Every action is recorded** — what was clicked, what was typed and what was on screen — in the task's history. Keep passwords, card numbers and one-time codes out of any automated flow. If a job needs them, do that step yourself and let the agent carry on. ### Ideas to automate this week - Rename and sort a folder of downloads into dated project folders. - Read three reports and write one summary of where they disagree. - Find every file in a project that still uses an old company name. - Copy figures from a desktop program with no export into a spreadsheet, with computer use. - Turn a finished post into follow-up tasks with recipes that run after you approve it. ### Frequently asked questions **Is letting an AI control my computer safe?** It depends on the brakes around it. In C5 it is off until an administrator turns it on; agents work only with programs you allowed, risky-looking clicks ask first, unattended runs need a time or step limit, and every action is recorded. Try it on something harmless first, and stay at the desk the first few times. **Can the AI work while I use my computer?** Yes, for most automation: file, web and integration work runs in the background while you carry on. Computer use is the exception — you and the agent share one pointer, so let it finish or press Stop rather than moving the mouse yourself. **Which apps can C5 automate on my computer?** Any service with an integration or an MCP server, through that connection. For a program with no other way in, computer use can work with it by sight, as long as you have allowed that program. The current integration list is in C5 under Settings › Integrations. **Does C5 support MCP servers?** Yes. Add one under Settings › Integrations with its command or address, then switch on only the abilities you want agents to have. MCP tools go through the same permission checks as everything else and are recorded the same way. **Can I stop an automation halfway?** Yes. Click Stop and the agent stops taking new steps right away. Stop is a brake, not an undo: a file already written stays written, and a message already sent stays sent. --- ## Remote control for the AI agent on your own computer Source: https://growther.si/ai-agent-remote-control AI agent remote control lets you use an agent running on your own computer from a phone, tablet or another computer. C5's Remote Control is encrypted on every hop, off until you turn it on at the machine, and admits only devices you have paired. Reach it over your own network, your own SSH with growther connect, or an address Growther.si issues. ### Three ways to reach your C5 Away from your desk, you still want to see what the agents did and answer what is waiting. Your agents, files and history stay on [a local AI agent](https://growther.si/local-ai-agent); Remote Control only gives you a way back to it. Pick the path that suits where you are and who you are happy to have in it. | Way in | How your device gets there | Third party in the path | | --- | --- | --- | | Your own network, or a name and certificate you supply | Straight to the machine | None | | Your own SSH connection, with growther connect | Through a channel your SSH already secures | None | | An address Growther.si issues (