A lot of privacy-oriented users have had the question: why is @ollama partly a cloud service a dependency for root?
The key lies not that a software is private by design, but rather can be used privately.
Our article on how to use @ollama anonymously gets you started on that.
Read it here:
https://t.co/ts0ljkOJgZ
Submissions page is open. Tonight in the $PALM Spaces #120, we will be browsing entries and how they were (with proof) created by using root.
Plus how you can make yours too. The best one wins 1,000 USD.
https://t.co/7q0okN1Ber
There is a better way to do what you do with @claudeai.
With no cost, full privacy and zero telemetry retained, let alone created.
What more, it is free until August 24th. 300 people have already shifted their repositories and skills over from Claude and Cursor to @rootsh, developed by $PALM.
Be one of them and pick better. Check it out at https://t.co/3s0MYGBnBF.
Note to all users: @ollama deepseek-v4-flash:cloud now automatically pulls the @deepseek_ai Flash-0731 model. It does not have to be configured separately.
1.0.2 now graphically shows the base model as well as its version, so you can keep track of which model and what version of it you are using.
The $1,000 Best App built with @rootsh competition is still on until August 15th. This feature might come in handy.
You can already try 1.0.2 if you are signed up for downloads on https://t.co/3s0MYGBnBF with a $PALM profile.
The website is getting a glow-up soon, so if there are rapid changes do not be alerted, it is still the same page.
We just made your response times drop by 99% to 3-6 seconds instead of minutes.
@rootsh now has Ask & Plan modes to reduce answer time to a few seconds and drop token usage by up to 90% (see the new response times from the video!)
You actually only need agentic power for agentic tasks. root now has three modes to support that.
- Agent: the agentic response model. Takes minutes for tasks that are complex & demanding, but does them really well and to finalization.
- Ask: fetches only the necessary info and processes to answer your question.
- Plan: Reads and studies instead of executing to build a plan for the Agentic mode to then put into action.
This update puts root at the same level as all other frontier-level harnesses in terms of token spend and usage as well as response times.
We just made your response times drop by 99% to 3-6 seconds instead of minutes.
@rootsh now has Ask & Plan modes to reduce answer time to a few seconds and drop token usage by up to 90% (see the new response times from the video!)
You actually only need agentic power for agentic tasks. root now has three modes to support that.
- Agent: the agentic response model. Takes minutes for tasks that are complex & demanding, but does them really well and to finalization.
- Ask: fetches only the necessary info and processes to answer your question.
- Plan: Reads and studies instead of executing to build a plan for the Agentic mode to then put into action.
This update puts root at the same level as all other frontier-level harnesses in terms of token spend and usage as well as response times.
Beta Feature: Full Avatar Customization 👤
You can now fully edit and customize the looks, sound, style and vibe of your Companion Avatar inside @rootsh.
Powered by local image generation and loadable profiles which you can customize at any time, you can now have an avatar that's not only 100% local and private, but:
- Fits to your persona and preferences,
- Is 100% customizable without restrictions,
- Stays only on your local device, with files portable anywhere,
- Has no guardrails as long as the model it is using has none (vision required).
In future versions, we will make it so that vision can be delegated to a guardrailed cloud model while the local input and output of text and observing can be delegated to a small-size local model.
More on the way. Built by $PALM.
Build an app at no cost and earn $1,000 ⚙️
The catch: you have to do it with @rootsh, our AI harness that matches up to @claudeai, @cursor_ai & all the others, but keeps everything private.
Rules are simple. You can only use root with any supported model, local or cloud. You have to use the Latest Stable version.
We recommend using the Templates that can get you started in under a minute on a beautiful & functioning app.
When you are done, we require you to send the Thinking logs securely without edits. The thinking logs have an embedded signature that proves their authenticity.
You do not have to do the app in one chat or sitting! You can build it over time, with different models, reiterating & going back, even rebuilding it. What matters is that it's done with root.
The $PALM Community is the backbone of the project and we wish to activate & rewards the amazing and resilient people who keep their trust in us. As such only those who hold over 1,000 $PALM are eligible. It does not matter when you started holding and we won't make that an issue.
The competition starts now (31st) and runs until August 15th. Good luck!
👉 https://t.co/3s0MYGBnBF
Created with @Kimi_Moonshot K3:cloud inside https://t.co/aBbJDBOo16 harness using the Builder (Blender + Unity) MCP.
Start building now — for free. Limited time only.
👉 https://t.co/aBbJDBOo16
⚙️ Kimi K3 is now supported for chat in Root, for both cloud and local Ollama installs.
$PALM will be offering a limited amount of testers free access in the coming days.
While out there, the K3 model is expensive to run in cloud and takes a toll on your hardware to run locally. We are at the top of solving this issue at the moment for millions of prospective users.
Open Weights
https://t.co/uLqkpA6auu
Ollama Service
https://t.co/4rYoWlYhXm
⚙️ Kimi K3 is now supported for chat in Root, for both cloud and local Ollama installs.
$PALM will be offering a limited amount of testers free access in the coming days.
While out there, the K3 model is expensive to run in cloud and takes a toll on your hardware to run locally. We are at the top of solving this issue at the moment for millions of prospective users.
Open Weights
https://t.co/uLqkpA6auu
Ollama Service
https://t.co/4rYoWlYhXm
Teaser: We are incorporating as a feature a private AI-powered browser within @rootsh to complement the deep research where you can do queries along with the AI and pass searched data.
While you do your work, you have a quick way to do private queries with a telemetry control layer blocking anything apart from your search queries from being sent over network, acting as an anonymous User Agent on the embedded browser.
You do the detailed query to a non-indexed site, AI will do the generic queries for research sources.
Along with this update the following change was pushed: chat texts are now by default the font of your selected theme. They were system font previously for clarity, it can be toggled in the settings. This looks better though.
root is a local AI harness with frontier-level models and all the tools you need. But there is more.
@rootsh is also an entity composed of people standing behind an idea.
That idea is privacy being a fundamental right not to be legislated over.
Read why we build.
The first article in our series, arguing that those in favour of legality outranking privacy were never in favour of it in first place.
With tips on what you can concretely do for your privacy, without permission for anyone. https://t.co/FGQC7S4qRb
The first article in our series, arguing that those in favour of legality outranking privacy were never in favour of it in first place.
With tips on what you can concretely do for your privacy, without permission for anyone. https://t.co/FGQC7S4qRb
Update: Root now has a custom, AI-native IDE integrated inside of it and supports running multiple tabs within one view.
No more IDE agent integration shopping and tab browsing - everything is one, in one single view.
Available on the local AI harness solving real issues, @rootsh.
You can now switch models on-the-go in @rootsh smoothly, as all models are pre-warmed on start-up.
In addition, chats now have a visual indicator what purpose they are serving - chatting, coding, researching, modeling with Builder, or clipping & creating with Creator Studio.
Changelog for the 0.0.8 Patch of 27th July:
- Settings menu is now visually adapted to quick usage, no more doomscrolling settings
- Added the Privacy Composer
- Merged 3D Builder into one unified "Builder" for both the Unity and Blender MCPs, allowing them to work together.
- Tens of quality changes to the installer, incl. -pre-selecting a theme and setting up some of the core settings before boot
Thanks to all the active testers for feedback!
The best way to get yourself or your organization compromised with AI is one secret key in prompt slip up.
You feed the AI the full code or a copy paste that includes your credentials. Then pray those will never be "out there".
root fixes this by recognizing and redacting them.
The most solidifying part of privacy is being in control. To enhance the apps utility, we thought we should add something generally useful that supports privacy.
And we did - no secret credentials can now pass through @rootsh to compromise yourself with in-built prevention.
Introducing as a Feature: Root Privacy Composer + Anti-Virus.
It is now impossible for you or anyone or your team to send secret keys, credentials, personal information and other privacy killers to be processed by AI. You can still send the secret keys as they are, but before shipping the prompt Root will redact every single credential. This means your repository and workspace remains clear of any compromising detail.
More than that, there is a simple antivirus feature now that can find any malware, spyware or resource eating apps directly on root. Root will also act on them the best it can to shut down any malicious runtimes, with the main category being unknown signature application memory use (malware).
Nothing is uploaded for scanning — patterns that are used for the privacy composer match only on your PC.
🛠 Root Version 0.0.8 is now available to download through the new installer.
You will now be prompted to accept our newly made Terms of Service and you can pre-select the theme on the application, overwriting the default selection.
Get 0.0.8 from here now:
https://t.co/UY88QFb1Bo
The @rootsh installer has been re-built to include dependencies.
Simpler install with one-click route to get everything needed to run the app on your PC in minutes.
The only current mandatory dependency is @ollama. Plugins may required dependencies such as MSOLAP or @github CLI to perform - they are installed with the plugin.
🟢 Live on https://t.co/3s0MYGBnBF
Ahead of the @rootsh Public Release, we want to demonstrate clearly where /root fits within the current AI landscape.
Harnesses are software layers that connect AI models to tools, files, memory, workflows, and interfaces so they can perform useful work in the style of @claudeai, @Cursor and @ChatGPTapp.
Simple work and chat tools like the web versions of the forementioned ones or @ollama provide an open runtime, and they can be either local or hosted, @VeniceAI as an example of the latter. Some local harnesses on the market, like Odyssey move close with the hosted ones to achieve the same result.
/root is being built to combine all these capabilities into one complete environment and exceed them. It is a full-suite harness with encryption and true privacy.
It brings frontier-level intelligence, agents that can edit files, run tools and ship work, a full creative and productivity suite, an integrated coding experience, complete telemetry control and an open model runtime you own. It is local and private by default, your codebase stays on your machine, and it can operate without ongoing token costs.
The goal is not to become another isolated AI tool, rather it is to deliver the complete private AI harness for coding, creativity, productivity and everything in between.
Privacy is a right, and so is proper access to performant AI without a dependency expense.
The Public Release is getting closer. $PALM is working.