Features & Usage
Learn how to make the most of your THOX.ai device's capabilities.
Popular Guides
AI-assisted coding
Match the model, client and workflow.
Client support
Inline completion, chat, keyboard shortcuts and project context are features of the client you configure. Verify them in that client; an API connection alone does not provide a THOX editor extension.
Review generated code
Use a small synthetic example first. Review suggestions, run relevant tests and check dependencies before using generated code in your project.
Choose an available model
Availability and hardware fit are separate checks.
Discover models
Use the model list returned by your selected endpoint. A model named in a catalog or roadmap may not be installed, served or available to your key.
Check fit
For local inference, confirm the exact artifact, quantization, runtime, memory and context requirements. Hosted models depend on your entitlement and service availability.
Compare results
Compare models on representative tasks using the same prompt and output limit. Parameter count alone does not establish quality or speed.
Chat and AI workspaces
Understand what each application sends and retains.
Choose a workflow
Open the application you intend to use and review its current availability and provider notice. Website demos and device runtimes have different capabilities.
Context and history
Only assume files or earlier messages are included when the application shows that behavior. Review its history, export and deletion controls before sharing information.
Actions and proposals
Generated text or an action preview does not authorize execution. The website AI demos do not control your devices.
Manage project context
Share only the material needed for the task.
Choose inputs
Select relevant files or a short excerpt. Remove secrets and confidential information before sending content to a hosted service.
Indexing and exclusions
Indexing and ignore-file support depend on the selected client. Check its settings; this website does not install a universal project indexer or .thoxignore handler.
APIs and integrations
Use the contract for your selected service.
Hosted inference
ThoxLLM Cloud exposes authenticated GET /v1/models and POST /v1/chat/completions at llm.thox.ai. Use Dashboard > API Keys if your account has the required entitlement.
Compatibility and limits
Chat compatibility does not imply support for every OpenAI endpoint, parameter, embedding model or tool. Check the service reference and returned error details.
Credentials
Keep API keys on the server or in the approved client credential store. Website sign-in, OAuth application registration and inference keys are distinct.
Measure performance
Benchmark the actual route and hardware.
Measure consistently
Record the model identifier, runtime, context length, output length and whether the model was already loaded. Separate first-token latency from generation speed.
Check the bottleneck
Compare network latency, model loading and inference time. A model-size threshold does not guarantee automatic acceleration or a particular speedup.
Sustained workloads
Follow hardware ventilation and operating limits. Recheck memory use and temperatures during a sustained test before increasing concurrency.