LiveContext Files

What is new

Every provider's current models are back, Gemini can call tools again, and linking to the cloud is open to every plan. v0.3.2 shipped a model catalog that followed what the cloud hosts, so an upgrade hid every Mistral, xAI, Qwen, Kimi, Z.ai and MiniMax model; this release ships a catalog curated for self-hosted installs.

Covers everything shipped since v0.3.2 (2026-09-26).

Before you update

• Any cloud plan can link again, Free included. v0.3.2 reserved the cloud link to paid plans. That is corrected: linking only requires a completed cloud onboarding (email confirmed), and what a link gives you for free is the model and skill catalog updates. Only what spends cloud money needs a paid plan: cloud models through the relay, the catalog relay and the cloud web search. A link suspended by v0.3.2 is restored with no re-link.
• 120 models, curated for self-hosted installs (the catalog shipped in v0.3.2 carried 119 and hid whole providers). Anything that referenced a model the catalog no longer carries runs on its replacement or the platform default instead of failing. Models you added yourself and edits you made to a model are kept.

Models

• The current models of every provider, among them:
• Anthropic: Claude Opus 5.5, Fable 5.1, Sonnet 5, Opus 5 and 4.x, Haiku 4.5.
• OpenAI: GPT-6 Astra, Sol and Luna, the GPT-5.6 family, GPT-5.5 and 5.4, the GPT Image models.
• Google: Gemini 3.1 Pro, Gemini 3.8 Flash (and 3.5 to 3.7 Flash), Gemini 3 Pro Image.
• Mistral: the current Large, Medium, Small, Ministral 14B / 8B / 3B, Codestral, Pixtral Large and the code models, through their -latest aliases.
• xAI: Grok 4.7, 4.6, 4.5 and 4.3. Qwen: Qwen3.8-Max, Qwen3.7-Plus, Qwen3.8-Flash, Qwen3.8-Omni-Flash.
• Kimi: K3, K2.7 Code, K2.6. Z.ai: GLM-5.3, GLM-5.3-Flash and the GLM-5 family. MiniMax: M3 and M2.7.
• DeepSeek: DeepSeek Flash, V4 Pro, Chat and Coder. TypeSafe: Jev for the Classify node.
• The Claude Code, Codex, Gemini CLI and Mistral Vibe bridges carry their current models too.
• Gemini can call tools again. Every Gemini conversation whose model called a tool failed on the next request: the tool result was sent in a format Gemini now refuses, and the signature Gemini attaches to each tool call was not sent back. Both are fixed, including several tool calls made in the same step.
• Using them: with your own key for the provider, as for any model. Through the cloud (linked installs in cloud mode), the cloud currently runs Anthropic, OpenAI, Google, DeepSeek and Mistral; Grok, Qwen, Kimi, GLM and MiniMax need your own key for now.

Workflows

• An Analysis tab next to Run. The latest epochs of a run side by side: outcome, work duration and cost per epoch, a grid of every node across epochs, and plain-language findings such as the node that keeps failing and its last error.
• Run logs show a stable ID. Each row is identified by where it ran (epoch, spawn, loop iteration, item: 21, 20.1, 20.0.2), so the same execution keeps its number across filters, pages and new arrivals. Epoch, spawn and iteration are always shown, first.
• A split followed by an aggregate no longer returns a single value instead of one per item when the run resumed on another server.

Chat and interface

• One clear error when a chat turn fails. The raw red strip above the conversation is gone; the error window says what happened (provider billing, rate limit, context too long, content blocked, model unavailable, network...) and what to do, in your language, with the raw text under "Technical details".
• The left sidebar animates its collapse and its mobile slide again.

Self-hosting

• A failing catalog download no longer repeats every 15 minutes. An install that could not apply a bundle downloaded it again and again (about 780 MB an hour on some). The model, skill and API catalog syncs now back off (10 minutes up to 6 hours) after a failure and remember it across restarts; "Sync now" still runs immediately.

Website

• /models describes models, not what the cloud hosts, and adds the newest ones (Claude Opus 5.5, GPT-6 Sol and Luna, Gemini 3.8 Flash, Grok 4.7, Qwen3.8-Flash, Kimi K2.7 Code, GLM-5.3-Flash and others).
• The footer lists the eight model families again (Claude, GPT, Gemini, Grok, Mistral, DeepSeek, Qwen, Kimi).

Cloud accounts

• Signing in with Google or GitHub works again. Since two-factor authentication arrived, a social login of an account without an authenticator app ended on an error page. It now signs in directly; an account with an authenticator app is still asked for its code.

This release also includes everything listed for v0.3.2.

Update an existing install

git pull
docker compose pull
docker compose up -d

The compose file pins this release's image tag, so run git pull first. Your data lives in Docker volumes and is kept across updates.

Fresh install

git clone https://github.com/livecontext-ai/livecontext-ce
cd livecontext-ce
docker compose up -d

Then open http://localhost:3000

Images

Pull from whichever registry you prefer, the digests are identical.

• ghcr.io/livecontext-ai/livecontext-ce:v0.3.3
• ghcr.io/livecontext-ai/livecontext-ce-frontend:v0.3.3
• ghcr.io/livecontext-ai/livecontext-ce-bridge:v0.3.3

• ghcr.io/livecontext-ai/livecontext-ce-screenshot-renderer:v0.3.3 (opt-in renderer profile)

• docker.io/livecontext/livecontext-ce:v0.3.3

• docker.io/livecontext/livecontext-ce-frontend:v0.3.3
• docker.io/livecontext/livecontext-ce-bridge:v0.3.3

• docker.io/livecontext/livecontext-ce-screenshot-renderer:v0.3.3 (opt-in renderer profile)

• quay.io/livecontext/livecontext-ce:v0.3.3

• quay.io/livecontext/livecontext-ce-frontend:v0.3.3
• quay.io/livecontext/livecontext-ce-bridge:v0.3.3
• quay.io/livecontext/livecontext-ce-screenshot-renderer:v0.3.3 (opt-in renderer profile)

Each image is also tagged v0.3, v0 and latest if you prefer to track a moving line rather than pin an exact release.