Four features, plus a latent bug fix.
llama.cpp backend. Claude Code only ever speaks the Anthropic Messages
API — confirmed empirically by pointing it at a logging server, which
received POST /v1/messages?beta=true. llama-server implements that
natively (verified in its README, alongside --port default 8080), so
this is a plain base-URL backend with no translation shim, the same
shape as Ollama. Its --api-key defaults to none, so the auth token is a
placeholder Claude Code requires and llama-server ignores.
Model alias fix. ANTHROPIC_DEFAULT_HAIKU_MODEL is documented as "also
used for background functionality", and Triple-C set none of the alias
vars. So on every custom-endpoint backend, Claude Code resolved `haiku`
to an Anthropic model id and sent it to a local server that does not
have it — background features failed silently. All four
ANTHROPIC_DEFAULT_{OPUS,SONNET,HAIKU,FABLE}_MODEL vars are now pinned to
the backend's configured model, with an optional Haiku override, and
blanked for Anthropic and Bedrock so those keep Claude Code's defaults.
The deprecated ANTHROPIC_SMALL_FAST_MODEL is never emitted. Existing
Ollama and OpenAI-Compatible containers are recreated once so the new
env reaches them; the snapshot is preserved.
Model gateway. Optional LiteLLM sibling container, off by default,
mirroring stt.rs — this is what makes real OpenAI usable, since
api.openai.com has no /v1/messages. Pinned to v1.96.0 by tag and digest:
the 1.82.7/1.82.8 malware was PyPI-only and never affected the official
images, which is precisely why this builds FROM the image rather than
pip-installing, but 1.84.0 is still the floor for proxy CVEs (API-key
SQLi, Host-header auth bypass, MCP auth bypass). Binds 0.0.0.0 because
project containers consume it, and therefore always sets a master_key —
LiteLLM without one accepts any key. The provider key lives in the OS
keychain and is uploaded into a volume, never an image layer or label.
URL relay. A container-side xdg-open/BROWSER shim opens URLs in the
host's browser. Uses an OSC sequence to /dev/tty rather than a printed
sentinel, because the shim usually runs as a grandchild of a process
capturing its children's output. Degrades to printing the URL when no
terminal is attached, so scheduled tasks do not hang. Only http/https,
with control characters rejected before new URL() — which strips
newlines, so java\nscript: would otherwise parse as javascript:. Nothing
auto-opens; the user confirms. The web terminal shows a tap-to-open
banner instead, since that browser may be a phone across a tunnel.
Browser view. A Project Home tab that watches and takes over the browser
Claude drives with Playwright, using Playwright's own dashboard. Zero
image cost — Playwright stays user-installed. It does not reuse the auth
bridge's PortForward, which binds an unauthenticated port: correct for a
throwaway OAuth listener, wrong for mouse and keyboard control of a
browser in a passwordless-sudo container. Instead a token-gated loopback
proxy checks Host, then token or a forbidden-header origin signal,
before a byte reaches the container. Host ports are confined to
47820..=47827 so CSP frame-src can enumerate them rather than widening
to a wildcard, with a test asserting the two agree.
188 frontend tests, 107 Rust tests, both builds clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
UI for the shared Claude token: a Settings section showing token state
with Authenticate and Revoke, an acquisition modal built on the shared
Modal (sign-in link handed to the host browser via the opener plugin,
plus the code input that answers `setup-token`'s stdin prompt — the flow
cannot complete without it), and a per-project opt-out toggle shown only
for the Anthropic backend.
Cancellation: acquire_claude_token previously had only two exits,
completion and a 15-minute timeout, and held the single-flight guard for
the whole time. Closing the dialog therefore locked the user out of
retrying for up to 15 minutes. Adds cancel_claude_token, backed by a
oneshot claimed and released in lockstep with the input guard, selected
on in the run loop so it wins the race and tears the exec down. The
dialog's Cancel now calls it and closes either way.
Also refreshes CLAUDE.md, which had drifted: it documented the deleted
ProjectCard, and asserted that new IPC commands need permission grants
in capabilities/default.json — they do not, that file covers plugin
commands only. Adds the conventions that would otherwise bite:
container_needs_recreation() is purely label-based and never diffs env,
so container-affecting state needs its own label; and #[serde(default)]
on a bool yields false regardless of intent.
Corrects the claim that Reset preserves credentials. Reset calls
remove_project_volumes, which deletes both the home and claude-config
volumes, so it wipes ~/.claude, the OAuth token, installed skills and
session transcripts.
84 frontend tests, 34 Rust tests, both builds clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adds first-class support for Claude Code CLI features (2.1.71-2.1.110):
- New ClaudeCodeSettings struct with per-project and global defaults for
TUI mode, effort level, focus mode, thinking summaries, session recap,
auto-scroll, env scrub, and 1-hour prompt caching
- Settings injected as env vars (CLAUDE_CODE_NO_FLICKER, etc.) and
~/.claude/settings.json entries via entrypoint.sh merge block
- New ClaudeCodeSettingsModal component for configuring settings
- Session naming support (-n flag passed to claude CLI, shown in tabs)
- Relaxed reserved prefix filter: CLAUDE_CODE_* env vars now allowed in
custom env vars UI for power users
- Global SSH key path, git name, and git email now used as fallbacks
when per-project values are not set, with UI in SettingsPanel
- Fingerprint-based change detection triggers container recreation when
Claude Code settings change
- Updated README, HOW-TO-USE, and CLAUDE.md documentation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Reflects that this backend works with any OpenAI API-compatible endpoint
(LiteLLM, OpenRouter, vLLM, text-generation-inference, LocalAI, etc.),
not just LiteLLM. Includes serde aliases for backward compatibility with
existing projects.json files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix serde deserialization error: TypeScript sent "lit_llm" but Rust expected "lite_llm"
- Rename AuthMode enum to Backend across Rust and TypeScript (with serde alias for backward compat)
- Add container image update checking via registry digest comparison
- Improve Settings page: fix image address display spacing, remove per-project auth section
- Update UI labels from "Auth" to "Backend" throughout
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add two new auth modes for projects alongside Anthropic and Bedrock:
- Ollama: connect to local or remote Ollama servers via ANTHROPIC_BASE_URL
- LiteLLM: connect through a LiteLLM proxy gateway to 100+ model providers
Both modes inject ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN env vars into
the container, with optional model override via ANTHROPIC_MODEL. LiteLLM
API keys are stored securely in the OS keychain. Config changes trigger
automatic container recreation via fingerprinting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
CLAUDE.md provides guidance for Claude Code instances working in this
repo (build commands, architecture overview, key conventions).
HOW-TO-USE.md is a user-facing guide covering prerequisites, setup,
all application features, and troubleshooting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
shadow-testandClaude Opus 5