HybridClaw v0.15.0
Release Date: April 29, 2026
The operator-surface release — disaster-recovery backup and restore, Tailscale Funnel, browser model switching, brand-voice guardrails, production Salesforce inspection, agent team structure, faster warm starts, and tighter local API auth.
Highlights
- Backup and restore for runtime homes — Added
hybridclaw backupandhybridclaw backup restorefor WAL-safe full-state archives with manifest validation before rehydrating~/.hybridclaw. - Tailscale Funnel and tunnel visibility — Added a managed Tailscale Funnel provider backed by encrypted
TS_AUTHKEYsecrets, plus admin tunnel status and public URL visibility alongside gateway controls. - Model switching in web chat — Added a composer model switcher with discovered provider catalogs, provider icons, model metadata, and active-session routing parity with local slash commands.
- Brand voice and production Salesforce workflows — Added the repo-shipped
brand-voiceoutput guard and skill, plus a fuller read-only Salesforce skill with metadata, SOQL, secret-placeholder handling, and eval coverage. - Agent organization and CV foundations — Added agent org-chart fields, restorable team-structure revisions, and chronological CV generation from observed skill history.
Changed
- Gateway health is less provider-coupled — Gateway status and health endpoints use cached provider checks instead of blocking on live model-provider probes.
- Usage and trajectory handling is more operational — Token usage writes are buffered and batch-flushed, while stored skill trajectories receive stronger PII, secret, and confidential-info redaction with retention controls.
- Runtime secret handling is stricter — Host and container runtimes share sensitive environment filtering and web-search credential injection so Brave, Perplexity, and Tavily keys flow through intended secret channels.
- Documentation is consolidated and complete — The legacy
docs/developmenttree was removed, and the command reference now inventories bothhybridclawCLI commands and local slash commands.
Fixed
- Local API auth enforcement — Loopback OpenAI-compatible API requests require
WEB_API_TOKENorGATEWAY_API_TOKENinstead of treating loopback address alone as authentication. - Gateway token persistence — Generated gateway tokens are persisted once, shared with local TUI/eval clients, and preserved across later config reloads.
- Ngrok reconnect noise — Tunnel reconnect errors are deduplicated by normalized cause, audit run ids are correlated, and tunnel health checks remain enabled by default.
- HybridAI model prefixes and context UI —
hybridai/...model ids resolve without noisy warnings, and the web chat context ring popover renders correctly.
Contributors
Core
- @furukama — release coordination, runtime/operator features, docs, command references, and reliability hardening
- @maxnoller — product and roadmap contributions included in this release train
All Contributors
Full Changelog: v0.14.0...v0.15.0