Hosted sessions fail closed and wake runners lazily
Roder 0.1.14 hardens hosted browser authentication, per-request policy, tenant isolation, thread-scoped MCP credentials, and remote-runner command execution for hosted and sandbox-backed workflows.
Roder CLI releases
Changelog
User-facing notes for the roder command-line interface. These entries are based on the GitHub releasesfor tags that start with roder/, then rewritten around practical impact for people running Roder in a terminal.
Latest0.1.14Read release
0.1.13: Lifecycle recovery makes interrupted runs inspectable
Roder 0.1.13 adds durable per-turn lifecycle records, restart reconciliation, bounded shutdown diagnostics, eval-loop persistence defaults, and clearer compaction summaries in the TUI and app-server stream.
0.1.12: Agent teams get Codex V2-style lifecycle semantics
Roder 0.1.12 makes Ultra-powered teammate workflows more durable, with canonical agent trees, reusable follow-up turns, mailbox-aware waiting, typed inter-agent messages, and richer app-server team results.
0.1.11: Codex GPT-5.6 models and Ultra mode land in Roder
Roder 0.1.11 updates the Codex provider catalog with GPT-5.6 Sol, Terra, and Luna, makes Sol the default Codex model, and adds Ultra as a first-class effort for bounded multi-agent work.
0.1.10: Grok 4.5 becomes the default xAI and SuperGrok model
Roder 0.1.10 adds Grok 4.5 to both the direct xAI API-key provider and SuperGrok OAuth, with a 500k context window and high reasoning by default.
0.1.9: Provider sign-in is more reliable on WSL
Roder 0.1.9 makes browser-based provider auth easier to recover when automatic browser launching fails, and adds a direct Kimi Code API-key login path.
0.1.8: Runner lifecycle controls now reach app-server and SDK clients
Roder 0.1.8 completes the runner-lifecycle rollout by registering pause, resume, detach, and rejoin in the authoritative app-server method manifest, then regenerating the JSON schema and TypeScript and Python client types.
0.1.7: Blaxel sandboxes keep remote coding sessions alive between CLI runs
Roder now ships a first-party Blaxel sandbox runner and a roder runners lifecycle CLI for pause, resume, detach, and rejoin workflows.
0.1.6: Agent-swarm mode turns one instruction into coordinated parallel agents
Roder now ships a native agent-swarm mode for bounded, rate-limit-aware parallel work with ordered results.
0.1.5: Live Claude Code sessions exit cleanly with one Ctrl+C
Roder now interrupts in-flight turns during TUI teardown before exiting, including provider sessions that have spawned a Claude Code subprocess.
0.1.4: Web-search provider selection is available directly from the terminal
The CLI settings surface can now manage external web-search routers such as Firecrawl, Tavily, Perplexity, Parallel, and Synthetic alongside hosted modes.
0.1.3: Synthetic joins the first-party inference providers
Roder adds a Synthetic inference provider with built-in model aliases, OpenAI-compatible chat-completions transport, model discovery, and setup guidance.
0.1.2: Fireworks and Kimi Code expand the default provider set
Roder adds Fireworks AI inference and first-class Kimi Code support, including OAuth for Kimi subscriptions, API-key fallback, catalog entries, and provider documentation.
0.1.1: Media generation, package installs, and ACP make the CLI more extensible
Roder adds first-party image generation providers, one-command package installation, and a stdio Agent Client Protocol adapter backed by the app-server runtime.