---
title: "HolaClaw v1.5.0"
description: "A more efficient way to browse the web, skills you can write or import yourself, and assistants that work on a schedule."
updated: 2026-08-11
canonical: https://holaclaw.ai/changelog/v1.5.0
---

# HolaClaw v1.5.0

Released August 11, 2026.

A more efficient way to browse the web, skills you can write or import yourself, and assistants that work on a schedule.

## Highlights

### A better way to browse

New assistants browse through the HolaClaw-authored browser-navigation skill instead of the assistant's slower bundled tool, reaching pages in fewer steps and with fewer failed actions. Your browser choice is unchanged.

### Skills

Give your assistant new abilities without waiting for an update. Write your own in a guided wizard, or import one from ClawHub or a git repository, reviewing what it contains before anything is installed.

### Scheduled tasks

Have your assistant do something on a recurring schedule: a morning briefing, a weekly summary, a nightly check. Create and manage schedules from the app or the command line, and choose where each result is delivered.

### Muse Glimmer 30B

A new local model built for agentic work, with vision, now in the catalog you pick from when you create an assistant. It runs entirely on your Mac, with no API key and no internet connection, and needs 32 GB of RAM.

## Changes

### Feature

- Browse the web through the new browser-navigation skill, which reaches pages in fewer steps than the assistant's bundled browser tool. The Skills page now confirms before switching browsing off.
- Schedule recurring tasks for an assistant, from the app or the command line, with a delivery target for each run.
- Connect an assistant to Slack, with direct-message pairing, channel allowlists, and an optional mention requirement.
- Create your own skills through a guided wizard, in the app or the command line, with a detail view for each one.
- Import skills from ClawHub or a git repository, reviewing what a skill contains before anything is installed.
- Browse a searchable catalog of skills, separate from the list of skills your assistant already has.
- Get new HolaClaw skills as soon as they are published, without waiting for a new assistant image.
- Stop a reply while it is streaming, with the send button doubling as a stop control.
- Continue setting up an assistant that was created but never signed in to a provider, instead of starting over with a new one.
- Gather a shareable troubleshooting bundle from the install-failure screen, saved to Downloads and revealed in Finder.
- Get a reminder to start a fresh conversation after adding or changing a skill, since a conversation only picks up skills when it starts.
- Write longer prompts in chat: the composer grows as you type and can expand into a full-height drafting area.
- Run Muse Glimmer 30B as a local model.
- Pick from an up-to-date model list when creating an assistant, covering the GPT-5.6 family, Gemini 3.6 and 3.5 Flash, Grok 4.5, and a refreshed OpenRouter lineup.
- Reach an assistant's gateway from your Mac with the new `holaclaw forward-gateway` command.

### Fix

- Fixed "New Session" in chat: it archived the conversation but left the assistant's session in place, so a newly installed skill could stay invisible to the chat.
- Fixed the model chosen when creating an assistant: requests could resolve to the provider's default model instead. On OpenRouter that meant a router picking (and billing for) a model you never chose.
- Fixed long tool-based turns, such as a browser task working through a page, being reported as "No response from assistant" moments before the reply arrived.
- Fixed restoring a configuration-only backup, which replaced the whole assistant directory and destroyed the workspace, provider sign-in, and session history the backup deliberately excluded.
- Fixed local models in the Mistral and Ministral family never replying.
- Fixed assistant tools installing with system privileges inside the VM; they now install as the assistant's own restricted user.

### Other

- Updated the bundled assistant to OpenClaw 2026.7.1-2, on a new base image.
- Updated the bundled local inference engine.
- Assistants now run with an explicit command-execution policy, so a turn can no longer stall on an approval prompt nothing could answer. The VM remains the isolation boundary.
- The app and the `holaclaw --version` command now report a build-identifying version and build date, which helps when reporting a problem.