The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Google Antigravity SDK local models can run inference on your own machine, either through the SDK-managed LiteRT runtime or through a separate local server with an OpenAI-compatible API. In those documented local paths, an API key and internet connection are not required for model inference. That does not make every Antigravity workflow automatically offline or guarantee that no information ever leaves the computer: cloud planners, authentication, and network-using tools are optional configurations with different data flows.
This FAQ explains the two local routes, what Google actually says about privacy, the hardware required for the documented Gemma checkpoint, setup details, hybrid operation, version changes, and practical limits.
As an Amazon Associate I earn from qualifying purchases.
What are the two ways to run a local model in the Antigravity SDK?
Antigravity documents two separate local-model paths. LiteRTAgentConfig runs a supported .litertlm checkpoint through an SDK-managed LiteRT runtime, while LocalOpenAIAgentConfig connects to a local server such as Ollama, LM Studio, or vLLM.
| Route | Runtime ownership | What you provide | Best fit |
|---|---|---|---|
LiteRTAgentConfig |
The SDK manages the LiteRT local runtime and loopback server. | A compatible .litertlm checkpoint and a backend choice such as GPU, NPU, or CPU. |
You want an SDK-integrated path and have a supported checkpoint and enough memory. |
LocalOpenAIAgentConfig |
You run and maintain the separate local model server. | The server’s model identifier and local base_url. |
You already use a local OpenAI-compatible endpoint or need a model/server combination outside LiteRT. |
These routes are not interchangeable. Google specifically says to use LiteRTAgentConfig when connecting to litert-lm serve; LocalOpenAIAgentConfig is for an external server exposing an OpenAI-compatible endpoint.
Popular on Amazon
- 【90 Degree Full Function USB C Adapter】Supports the latest USB4 standard for up to 40Gbps of bandwidth, 8K video support, and up to 240W Fast charging; Support USB PD 3.1 Extended Power Range (EPR) including 140W, 180W and 240W. Your productivity is easily doubled. *Warm Tip: For the best fit, please make sure the USB-C port is fully exposed; thick or bulky cases may affect compatibility.
- 【Reliable Quality】Made from high-quality, scratch-resistant aluminum alloy, delicate metal frame. get excellent wear resistance and heat dissipation. Non-Slip Design for easier insertion and removal.
- 【Small and portable】Ultra thin size, ultra compact design, a petite physique that matches the minimalist design of your modern laptops. Coming with a slim figure that won't block the port next to the one it's plugged into.
- 【Compatibility】Widely work with most usb c, thunderbolt 4/3 devices. Compatibilit for ROG Ally, Steam Deck, Switch, MacBook Pro, Quest 3 VR headset, Surface Book 2, iPhone 18/17/16/15, Samsung Galaxy S23/S22/S21, Note 20/10, Google Pixel, ChromeBook, etc.
- 【Excellence Customer Support】If you experience any issues or cable failure, simply reach out to our dedicated customer service team. We provide a prompt response within 18 hours, providing you with the necessary technical support to resolve any concerns.
In both cases, “local” describes where model inference runs. It does not automatically describe every tool, planner, telemetry setting, or file operation in the surrounding agent workflow.
Can I run Antigravity SDK local models offline?
Yes, for the documented local-inference paths, Google says agents can run with local models without an API key or internet connection. The Google Developers Blog describes the capability as enabling agentic assistance with local models “completely offline,” and the technical documentation distinguishes local models from cloud authentication requirements.
Offline operation still requires preparation. You must install the SDK and runtime packages, obtain the checkpoint or local server model while connected, and make sure any dependencies are available on the machine. Once those pieces are present, inference can stay on the device.
Do not interpret this as a guarantee that an entire application process is network-isolated. A workflow can still contact the network if you configure a cloud model, use a cloud planner, authenticate against a hosted service, or invoke a tool that reads a remote resource. Review the agent’s tools and policies separately from the model configuration.
Does local model use send my code to Google?
Google documents local inference as keeping code and requests on the machine when the agent uses a local model. That is a description of the local execution path, not a universal promise that every Antigravity configuration is data-free or isolated from Google services.
Google’s hybrid example makes the distinction concrete: the cloud planner received filenames and task descriptions while source code remained local. Those details describe that recorded demonstration, not a blanket rule for every hybrid workflow, tool, or future release. If your configuration sends prompts, metadata, filenames, logs, or planning context to a cloud component, that information follows the cloud component’s data policies.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallAlso popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
The product FAQ directs readers to the Antigravity Terms of Service and FAQ for the broader data-collection stance and says data collection can be turned off from the Settings panel. That opt-out is a product setting; it should not be presented as a detailed audit of every SDK-local runtime data path.
For a sensitive codebase, establish a written data-flow map before enabling hybrid features. Identify which model sees source text, which component sees filenames or task descriptions, whether tools can make network requests, where logs are written, and which workspace policy controls those actions.
What hardware do I need for Gemma 4 26B A4B?
For the Gemma 4 26B A4B checkpoint used in Google’s documentation, the recommendation is at least 24 GB of VRAM or unified memory, and the download is approximately 16.8 GB. Those figures apply to that checkpoint example; they are not an SDK-wide minimum for every local model.
A dedicated 24 GB graphics card is therefore one possible configuration, not a mandatory purchase. A machine with sufficient unified memory can also meet the documented recommendation. Smaller checkpoints may have lower requirements, but the supplied Antigravity material does not provide a universal memory calculator or performance table for other models.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsLeave room for the operating system, the Python environment, the model files, and runtime overhead. A machine that barely meets the checkpoint’s memory recommendation may still encounter loading failures or excessive swapping if other applications consume memory. Select the LiteRT backend that your hardware supports—GPU, NPU, or CPU—and verify that the checkpoint is compatible with that backend.
How do I set up a LiteRT local model?
You need a Python environment, the Antigravity SDK package, the LiteRT tooling, enough disk space for the checkpoint, and hardware compatible with the selected model. For the documented Gemma example, plan for roughly 16.8 GB just for the download and at least 24 GB of VRAM or unified memory.
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
- Install the packages. In the environment used by your project, install the two documented packages:
pip install google-antigravity litert-lm. - Import the checkpoint. Use the documented
litert-lm importcommand to convert or install the supported source checkpoint into a.litertlmfile. The exact arguments depend on the checkpoint source and the current LiteRT documentation, so copy the command for the model you selected rather than substituting an arbitrary filename. - Choose a backend. LiteRT configuration exposes GPU, NPU, and CPU choices. Use a backend supported by both your operating system and the checkpoint; a nominally available accelerator is not proof that every model can run on it.
- Pass the full checkpoint path. Set the checkpoint path in
LiteRTAgentConfig. The Antigravity docs warn that~is not expanded automatically, so provide an absolute path or expand it in your own Python code before constructing the configuration. - Create the agent with the LiteRT configuration. Select the model and backend in
LiteRTAgentConfig, then start the agent using the normal SDK flow. The SDK launches its local runtime and loopback service for the checkpoint. - Check the first startup. The initial run may spend time loading a large file and initializing the accelerator. Confirm that the selected backend is detected and that the process can read the checkpoint before diagnosing prompt or tool problems.
LiteRT configuration automatically applies a lightweight preset. According to Google’s documentation, it narrows built-in tools to core coding operations, trims system instructions for smaller context windows, and applies a default compaction threshold sized for a 64k LiteRT KV cache. Those defaults help fit a local runtime but can change the available context and tools compared with a cloud configuration.
How do I use Ollama, LM Studio, or vLLM with Antigravity?
Use the external-server route when the model is already served by a local OpenAI-compatible endpoint. Start the server separately, confirm its local URL and model identifier, then configure LocalOpenAIAgentConfig with that identifier and the endpoint’s base_url.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Install and start one server. Set up Ollama, LM Studio, or vLLM according to that project’s instructions and download a model locally.
- Confirm the endpoint. Record the server’s local base URL and the exact model name it reports. A mismatch between the configured identifier and the server’s name is a common cause of “model not found” errors.
- Configure Antigravity. Pass the model identifier and local
base_urltoLocalOpenAIAgentConfig. Keep this configuration separate fromLiteRTAgentConfig; an external OpenAI-compatible server is not the LiteRT loopback service. - Run a minimal request first. Start with a short prompt that does not invoke tools. If that succeeds, add workspace tools and policies one at a time so connection errors can be distinguished from tool-permission errors.
- Keep the server running. The SDK depends on the separate process. Stopping it, changing its port, or unloading the model will interrupt subsequent agent calls even though the Antigravity Python process is still running.
This design separates model serving from the SDK. You can update, replace, or tune the local server independently, but you also own its memory settings, port, model files, logs, and access controls.
Should I choose LiteRT or an external local server?
Choose LiteRT when you want Google’s documented .litertlm path and SDK-managed runtime setup. Choose an external server when you already have a compatible local endpoint or need to manage model serving outside the SDK.
| Question | Prefer LiteRT | Prefer an external server |
|---|---|---|
| Who manages the runtime? | You want the SDK to manage the local LiteRT service. | You want an existing server process to remain independent. |
| What model format is available? | You have a supported .litertlm checkpoint. |
Your model is exposed through an OpenAI-compatible API. |
| What setup is simplest? | One SDK configuration after importing the checkpoint. | Server installation and lifecycle management are acceptable. |
| What hardware matters? | LiteRT backend support plus the checkpoint’s memory requirement. | The server’s own model, quantization, accelerator, and memory requirements. |
Neither route is automatically faster or more capable. The supplied Google material does not provide an independent local-versus-cloud speed, quality, or cost benchmark, so select based on compatibility, control, and data-flow requirements rather than an assumed performance advantage.
Can Antigravity combine local and cloud models?
Yes. A hybrid workflow can assign a cloud model to a planning role while local models perform on-device work. This can preserve local handling of source code while still using cloud reasoning for selected planning information, but it requires cloud authentication and a deliberate decision about what context is sent.
Also popular on Amazon
- Portable and powerful USB-C HUB: BENFEI USB Type-C HUB, with super-soft and knot-free silicone woven design cable, meets most mobile office needs. Compact, lightweight, stylish, and powerful portable USB C Hub equipped with 1 x HDMI port, 1 x 100W charging, and 3 x USB ports. 18-month warranty, 24-hour response, to ensure you feel at ease when using our product.
- Design centered on comfort and reliability: Thanks to BENFEI's end-to-end in-house cable production capability, in-house PCBA and assembly capability, using the industry's most advanced silicone woven design and process, 20cm cable in length, no knots, super-soft, the HUB is easy to use in all scenarios: laptop, tablet, stand etc. Super-soft, 25000+ life cycles, to meet your daily carrying and office needs.
- 100W Charging: Support up to 90W USB C pass-through charging via Type-C port to keep your laptop powered. 10W is reserved for other interface operations. No data and video function on the Type-C port.
- 4K HDMI Display: The HDMI port supports media display at resolutions up to 4K 30Hz, keeping every incredible moment detailed and ultra vivid. Please note that the C port of the Host device needs to support video output.
- Transfer Files in Seconds: Transfer files and from your laptop at speeds up to 10 Gbps with USB A 3.2 port. Extra 2 USB A 2.0 ports are perfectly for your keyboards and mouse.
In Google’s recorded demonstration, 3,322 tokens—97.2% of the example’s tokens—ran locally, while the cloud planner used 95 tokens. These are figures from Google’s example, not an independently validated benchmark or a typical result for every task.
Hybrid mode is therefore not “offline” for the entire workflow. If the planner is cloud-hosted, the process needs network access and the planner receives whatever context the configuration supplies. Treat the example’s separation of filenames and task descriptions from source code as one implementation pattern, then inspect your own prompts, tool adapters, and logging behavior.
What are the main limitations of Antigravity local models?
- Checkpoint compatibility: LiteRT requires a supported
.litertlmcheckpoint. A model that runs in another server is not automatically usable through LiteRT. - Memory and storage: Large checkpoints consume substantial disk and accelerator memory. Gemma 4 26B A4B’s approximately 16.8 GB download and 24 GB memory recommendation illustrate the scale.
- Context constraints: The LiteRT lightweight preset trims instructions and uses a compaction threshold sized for a 64k KV cache. Long tasks may be compacted sooner than you expect.
- Tool scope: The lightweight preset narrows built-in tools to core coding operations. A local model does not automatically grant every cloud workflow capability.
- External-server availability: With
LocalOpenAIAgentConfig, the server must be running, reachable at the configured URL, and serving the named model. - Network ambiguity: Local inference can be offline, but cloud planners, hosted authentication, and network-enabled tools can still communicate externally.
- Policy risk: Workspace and tool permissions determine what the agent can read or change. Review them for least privilege; do not copy broad permissions such as an example’s
policy.allow_all()into production without understanding the consequences.
What changed in recent Antigravity SDK versions?
The changelog records first-class local-model support in version 0.1.18, dated September 21, 2026. The latest reviewed entry is version 0.1.23, dated October 26, 2026, which adds multi-turn conversation caching for LiteRT models and KV-cache reuse for local LiteRT conversations.
These dates describe the entries reviewed in the changelog, not a promised release schedule or support lifetime. Local-model behavior can change as the SDK, LiteRT tooling, model formats, and backend integrations evolve. Check the current SDK changelog and the local-model documentation before upgrading a production environment.
Free tools Windows power users keep installed
One-click scans. No signup required.
How do I troubleshoot a local-model setup?
Start by identifying whether the failure is in the checkpoint, the runtime, the server connection, or an agent policy. The following order avoids changing several variables at once.
- Verify installation: Confirm that the Python environment contains both
google-antigravityandlitert-lm. - Verify the file: Check that the
.litertlmpath is absolute and readable. Replace an unexpanded~with the fully expanded path. - Verify memory: Compare available VRAM or unified memory with the selected checkpoint’s documented recommendation and close competing workloads.
- Verify backend selection: Try the backend supported by your hardware and model. If accelerator initialization fails, test CPU only as a diagnostic, not as evidence of acceptable production performance.
- Verify external endpoints: For Ollama, LM Studio, or vLLM, confirm the server is running, the port is correct, and the configured model identifier exactly matches the server.
- Verify policy and tools: If a simple prompt works but a file operation fails, inspect workspace permissions and the lightweight preset’s reduced tool set rather than reinstalling the model.
- Verify network assumptions: If a supposedly offline workflow makes a connection attempt, look for a cloud planner, authentication setting, telemetry option, or network-using tool in the configuration.
What should I check before deploying local inference?
Document the exact SDK version, checkpoint filename and hash if your organization tracks artifacts, backend, memory available, server URL, model identifier, enabled tools, and workspace policy. Decide whether logs or prompts may contain source code and where those logs are stored.
Also popular on Amazon
- 【Read before You Buy】 Note: This adapter is USB 2.0 version and cannot transmit HDM, VGA, or other video signals. Nevertheless, they are fully capable of charging devices,duplicating files, and can be seamlessly integrated with USB-C earphones, wireless network adapters, and SD card adapters.
- 【Easy to Use and Safe】Thanks to modern manufacturing techniques, we've made these adapters sufficiently compact so that they can remain inserted in your device without noticeable protrusion or inconvenience. This means you don't have to constantly plug and unplug them, reducing the likelihood of wear and misalignment, and making them more convenient to carry around.
- 【Reliable Quality】Engineered for enhanced durability, durable aluminum alloy shell material,get excellent wear resistance and heat dissipation,the male USB C and female USB connectors & the male USB and female USB C connectors are designed to be plugged and unplugged up to 20000 times. Provides data transfer, device charging, memory space expansion, and additional computer peripherals.
- 【Think more for you】Do you need to connect a USB-C-enabled device to an odd device that only has USB-A ports? Then these USB-A male to USB-C female adapters are made for you.Simply insert the adapter into the USB-A port on the device, and you can instantly connect your USB-C device to the device that originally didn't have the correct connection port.
- 【Trusted xiwxi Quality and Service】As a responsible brand seller, you will receive xiwxi's 12-month Quality assurance and lifetime friendly customer service, you're not just buying a product,you're investing in peace of mind with all-round quality and exceptional customer service.
Run a small, non-sensitive smoke test after every upgrade, then test the actual tool permissions in a disposable workspace. Keep a rollback copy of the previous SDK and checkpoint because the changelog does not promise a fixed update cadence or long-term compatibility.
Frequently Asked Questions
Can I use a local model without creating a Gemini API key?
Yes. Google’s authentication documentation says the documented local-model paths do not require an API key. You still need credentials if you add a cloud model, cloud planner, or another hosted service to the workflow.
Recommended Free Tools
Will a local Antigravity agent work while the computer has no GPU?
The LiteRT configuration lists CPU as a backend option, so a GPU is not universally mandatory. Suitability depends on the checkpoint’s compatibility and memory needs; the Gemma 4 26B A4B recommendation is still at least 24 GB of VRAM or unified memory.
Where are local-model conversations stored?
The supplied documentation does not define one universal storage location for prompts, caches, or logs across LiteRT and external servers. Check the SDK configuration, the local server’s settings, and your operating system’s application-data directories before treating a workflow as fully private.
Can I switch from LiteRT to Ollama without changing my model files?
Not necessarily. LiteRT expects a supported .litertlm checkpoint, while an external server accepts the formats and models that server supports. You may need a separate download or conversion, and the model identifier must be configured again.

