Sign in to continue.
Your chats are stored encrypted. Your password never leaves this device.
Your events, iCal subscriptions and CalDAV calendars in one place.
Local coding agent integration.
Config generator
Central management for models, system, language, and users.
The overview is already available. Settings are loading in the background.
Review the backup configuration and create a fresh snapshot.
Start with a task. Advanced settings stay out of the way until you need them.
Basic application settings.
Reopen the guided assistant or preview optional model presets. Your normal configuration stays under Models, Response modes, and Prompt Studio.
Walk through provider, model, response modes, prompt, and speech in four focused steps.
These are suggested starting values, not a provider list. Every OpenAI-compatible API can be configured without a template.
A provider hosts an OpenAI-compatible API. Add the models available through that connection underneath it.
Connection details belong to the provider; display names and model IDs belong to its models.
Optional OpenAI-compatible /v1/embeddings endpoint that adds semantic ranking to memory search. Leave the URL empty to keep keyword-only retrieval. Changes apply immediately without a restart.
These are the understandable choices users see—not model-specific implementation details.
Names and descriptions appear in the composer. A mode can force a specific model.
Choose a model, decide how ChatTLLX should assemble its instructions, and write the assistant behavior in plain language.
ChatTLLX adds its maintained tool, file, and artifact instructions around your prompt.
The preview contains the full generated prompt for this language and the capabilities configured on this instance. Individual turns may omit unavailable or user-disabled sections. Editing creates an exact override.
When customized, this text is sent exactly in Managed mode. Raw mode continues to send only the base prompt above.
These texts are sent beside the system prompt. Open only the tools you want to tune; names and schema types stay stable.
Most instances do not need these. Open a mode only when its behavior must differ from the model defaults above.
Choose where voice recordings are transcribed. Users only see the microphone when transcription is available.
You can change this later without touching Compose files.
The model is downloaded on first use and kept in the persistent model cache.
Changing the model needs no container restart. The next transcription loads the selected model.
ChatTLLX sends a standard multipart request to /v1/audio/transcriptions.
The microphone is hidden for users and the optional local Whisper container stays stopped.
These limits apply to both local and upstream transcription.
Visualizations, global tool switches, and API access.
Encrypted, versioned files of any type per chat and optionally per user. Model access requires an isolated runner.
Dedicated Linux containers for coding, tests, document processing, and general command-line work.
Use a dedicated Proxmox API token. ChatTLLX never needs SSH or host-shell access.
A signed, secret-free Debian LXC image with the guest agent and document, coding, and diagnostic tools preinstalled. Each installation supplies its own certificates and keys.
The public image never contains your Proxmox token, guest master, or private TLS key.Download every release file, then verify both signatures and the rootfs checksum before importing anything.
On the Proxmox HOST, run the verified script as root with an unused base CT ID, clone storage, bridge and optional VLAN. It downloads and verifies the signed image, creates a sealed, stopped base CT, and prints one-time enrollment JSON. Nothing needs to be typed inside a CT. Requires minisign on the host.
The host setup script automatically personalizes and seals the CT, then prints the enrollment JSON once. Keep the output private.
Paste the complete one-line JSON output. It fills only the Guest CA, Guest Master, and verified URL pattern.
Enter the stopped base CT ID and clone settings below, save, then run Test configuration. The host setup has already sealed and stopped the base CT.
Only the endpoint and dedicated API token are required.
Choose the stopped Agent-Ready base CT and where full clones should live.
Inactive chat containers stop automatically, remain assigned to their chat, and restart when that chat is opened.
A container assigned to one chat is never reused for another chat. Deleting the chat deletes its container first.
Reserve enough IDs for the warm pool and retained chat containers. ChatTLLX touches only journal-owned CTs.
Install verified releases with an encrypted backup, health checks, and automatic rollback.
Web updates require the restricted ChatTLLX host operator installed by the Compose installer.
Manage reusable skills.
Create, edit, and enable skills.
System health checks.
Checks security-sensitive configuration and service health without reading user content.
Loading diagnostics…
Encrypted backups.
Create encrypted application snapshots on a reliable schedule.
Create and restore encrypted application snapshots.
No backups yet.
Aggregated token usage for the entire system.
Input is the complete prompt. Cache reads are a subset of input; reasoning is a subset of output.
Only system-wide aggregates are returned. Account identities and individual usage remain private.
Review and edit the installed interface languages.
Manage users.
0 Users
Customize TLLX to your needs.