Aura

v1.2.0 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 22d AI Agents & Assistants

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

ai ai-agents deepseek desktop-app developer-tools llm

+4 more

openrouter pair-programming pyside6 python

Affected surfaces

auth

ReleasePort's take

Light signal

editorial:auto 13d

Worker 1.2.0 introduces a two-pane UI layout with live-streaming code tabs and pluggable CLI backends for Gemini, Claude Code, and Codex. Code display remains read-only and tool-call support is currently limited to the default backend.

Why it matters: The two-pane layout improves workspace organization. Multiple backends enable AI platform choice; however, read-only code display and limited tool-call support constrain integration. Evaluate backend compatibility in dev before upgrading.

Summary

AI summary

Worker UI adopts a two-pane layout and adds pluggable CLI backends (Gemini, Claude Code, Codex).

Changes in this release

Type	Severity	Summary	CVE
Security	Medium	Aura does not store credentials; tokens remain in tool config Aura does not store credentials; tokens remain in tool config Source: llm_adapter@2026-05-21 Confidence: high	—
Feature
Feature	Medium	Worker UI replaced single feed with two-pane layout Worker UI replaced single feed with two-pane layout Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Top pane displays code tabs with live streaming effect Top pane displays code tabs with live streaming effect Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Bottom pane shows reasoning, task list, terminal in tabs Bottom pane shows reasoning, task list, terminal in tabs Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Workspace automatically clears temporary tabs on task completion Workspace automatically clears temporary tabs on task completion Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Code tabs are read-only; direct editing not permitted Code tabs are read-only; direct editing not permitted Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Hook-based backend interface enables external CLI tool execution Hook-based backend interface enables external CLI tool execution Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	CLI backends include Connect Account button for authentication CLI backends include Connect Account button for authentication Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Gemini CLI backend available for sandboxed code generation Gemini CLI backend available for sandboxed code generation Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	Claude Code CLI backend available for sandboxed code generation Claude Code CLI backend available for sandboxed code generation Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	OpenAI Codex CLI backend available for sandboxed code generation OpenAI Codex CLI backend available for sandboxed code generation Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	Integrated terminal enables browser device authorization flow Integrated terminal enables browser device authorization flow Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	`generate_code` hook can be fulfilled by default API backend or registered CLI backends `generate_code` hook can be fulfilled by default API backend or registered CLI backends Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Feature	Medium	Gemini CLI backend uses `gcloud ai models predict` in sandbox and streams JSON responses Gemini CLI backend uses `gcloud ai models predict` in sandbox and streams JSON responses Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Feature	Medium	Claude Code CLI backend invokes Anthropic's `claude` CLI with tool‑use and filesystem support Claude Code CLI backend invokes Anthropic's `claude` CLI with tool‑use and filesystem support Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Feature	Medium	Codex CLI backend invokes OpenAI's `codex` CLI for sandboxed execution using existing credentials Codex CLI backend invokes OpenAI's `codex` CLI for sandboxed execution using existing credentials Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Feature	Medium	Pressing the Connect Account button opens an integrated terminal to run the tool's login command Pressing the Connect Account button opens an integrated terminal to run the tool's login command Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Performance	Medium	Worker file streaming smoothed for line-by-line delivery Worker file streaming smoothed for line-by-line delivery Source: llm_adapter@2026-05-21 Confidence: low	—
Performance	Medium	UI responsiveness improved during agent task execution UI responsiveness improved during agent task execution Source: llm_adapter@2026-05-21 Confidence: low	—
Deprecation	Medium	Tool-call support currently limited to default API backend Tool-call support currently limited to default API backend Source: llm_adapter@2026-05-21 Confidence: high	—

Full changelog

Worker UI: Two-Pane Workspace

Replaced the single chronological feed with a dual-pane layout.
Top pane displays code tabs with a live streaming effect during file generation.
Bottom pane separates reasoning plan, task list, and terminal output into distinct tabs.
Workspace automatically clears all temporary code and terminal tabs when the Worker completes a task.
Code tabs are read-only; direct editing is not permitted in this pane.

Pluggable CLI Agent Backends

Introduced a hook-based backend interface that allows the Worker to execute code generation via external CLI tools instead of a direct model API call.
The generate_code hook can be fulfilled by the default API backend or by registered CLI backends.
New backends are implemented as thin wrappers that format a spec, execute a CLI command, and parse the response stream.

Integrated CLI Backends

Three external CLI agents are now available as Worker backends:

Gemini CLI – uses gcloud ai models predict within the sandbox executor; parses JSON responses with multiple fallback paths; streams content as ContentDelta events.
Claude Code – invokes Anthropic’s claude CLI; leverages its tool-use and filesystem capabilities inside the Aura sandbox.
Codex CLI – invokes OpenAI’s codex CLI for sandboxed agentic execution; uses existing OpenAI credentials.

All CLI backends run via the existing SandboxExecutor. Responses are parsed and streamed to the Worker code pane. Tool-call support is currently limited to the default API backend; future updates may extend this to CLI backends.

Authentication

Each CLI backend provides a “Connect Account” button in the backend settings.
Pressing the button opens an integrated terminal inside Aura and runs the tool’s official login command (e.g., gcloud auth login, claude login, codex login).
The user completes the browser-based device authorization flow; Aura monitors the process exit status.
On success, the backend status updates to indicate an active connection.
Aura does not store or manage credentials; all authentication tokens remain in the tool’s default configuration files.