v0.18.1 : Update notes (2026/09/26)
Thank you for using Blaze!
v0.18.1 brings viewing and directly instructing sub-agents, tags for messages and threads with a search palette, adding Library items from Settings, support for the GPT-6 family and Opus 5.5, and improved Salesforce connections on corporate proxy networks.
1. You can now follow sub-agents’ work and instruct them directly
Click a sub-agent row in the conversation timeline to open a tab in the browser pane that shows the sub-agent’s progress in the same format as the conversation screen.
- You can send messages to the sub-agent directly from the tab’s input field. Messages sent while it is working are picked up in its next step, and messages sent after it finishes make it continue the work. The result is also passed back to the main conversation.
- From the tab’s input field, you can stop just that sub-agent or change the model it uses.
- The progress of finished sub-agents is saved, so you can continue them even after restarting the app.
- The sub-agent row in the timeline shows the beginning of the instructions Blaze gave it and its progress. When work is delegated to OpenAI Codex or Claude Code, you can also see the instructions that were passed.
We also made the following improvements:
- In Approval Mode, operations performed by a sub-agent now appear as approval cards in the main conversation, labeled as a request from that sub-agent. If you reject one, only that operation is canceled and the work continues.
- Sub-agents can now be used in Requirement Mode as well.
- You can now choose the “Default sub-agent model” in the “AI” section of Settings > General. The default is “Same as conversation.” If you specify a model in chat, that model is used instead.
- Research sub-agents can now look up data in your Salesforce org as part of their investigation.
See Subagents for details.
2. You can now tag messages and threads and find them later
- Hover over a message and choose “Tags” from the menu to tag it. Tags appear under the message as
#tag-name. You can also tag messages in past conversations. - You can also tag threads from “Tags” in the thread’s ”…” menu.
- Press ⌘K (Ctrl+K on Windows) to open the search palette, then type
#followed by a tag name to jump to tagged messages and threads. - From the search palette you can also search conversations, start a new conversation, and open Settings. Conversation search, previously in the sidebar, has moved into this search palette.
3. You can now add Skills and prompts from your Library in Settings
We added a “Library” tab to Settings. You can search and review the Skills and prompts shared to your organization’s library, and add or update them directly in the Blaze app.
- You can also ask Blaze in chat to add Skills or prompts from the library.
- The confirmation screen for sharing to the library and the screen for adding from the library have been simplified. When the action completes, the screen closes and a notification lets you know.
- When you update a prompt added from the library, any local edits are overwritten by the library version.
4. GPT-6 Astra / Sol / Luna and Opus 5.5 are now available
- You can now select GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, and Opus 5.5 from Blaze’s model selector.
- GPT-6 Astra and Opus 5.5 consume more credits. Please check your plan and your organization’s model usage settings.
- Opus 5 / Opus 4.8 / Fable 5 / GPT-5.5 / GPT-5.6 Sol / GPT-5.6 Luna have been removed from the model options. If you had one of these selected, it is automatically switched to the following model:
- Opus 5 / Opus 4.8 → Opus 5.5
- Fable 5 → Fable 5.1
- GPT-5.5 / GPT-5.6 Sol → GPT-6 Sol
- GPT-5.6 Luna → GPT-6 Luna
5. Salesforce connections now work on corporate proxy networks
Fixed an issue where logging in to Salesforce, or retrieving and deploying metadata, could fail on networks that route traffic through the proxy configured on your computer (including automatic configuration scripts).
- If the proxy asks for authentication when you log in to Salesforce, Blaze now shows “Proxy authentication is required” and suggests checking with your network administrator.
6. You can now attach images with Blazer 1 and keep the conversation going
When you attach an image with Blazer 1, an image-capable model reads the image, and Blazer 1 answers based on the result. The conversation continues on Blazer 1.
- You can follow the image reading progress in the conversation timeline.
7. Salesforce credentials are no longer kept in conversation history
When the AI ran commands such as sf org display, the Salesforce access token was saved as-is in the conversation history and sent to the AI.
Access tokens, refresh tokens, and auth URLs are now replaced with [REDACTED] in the displayed output, the conversation history, and what is sent to the AI.
8. Improvements to how Skills behave
- A Skill you have used now stays active for the rest of the conversation.
/browser-testkeeps testing with the browser even after you stop and send a follow-up, or restart the app. - Fixed an issue where some tools and MCP became unavailable while a Skill was in use, which could keep Blaze from working correctly.
- Removed items that have no effect in Blaze (fork execution, agent, model, Hooks, and allowed tools) from the Skill details in Settings > Skills.
9. Chat and UI improvements
- The AI now prioritizes your requests over each tool’s default usage guidelines.
- Files now appear faster after you open them.
- In Approval Mode and Requirement Mode, the To-Do card and message queue borders now use the mode color as well.
- You can now insert a line break with Shift+Enter when writing a free-text answer to a question from the AI.
- Changed the icon of the History and Downloads pages opened in the built-in browser.
10. Bug fixes
- Fixed an issue where the Microsoft Store version could stop launching.
- Fixed an issue where the selected row in the Settings > Files list had no background color.
- Fixed an issue in the Settings device list where a failure to load or delete devices showed no error and could make it look as if there were no devices.