Blaze

v0.5.0 : Update notes (2026/04/11)

💡 Thank you for using Blaze!


1. Force tool — Salesforce metadata in one shot

We added a new Force tool that has Blaze build Salesforce metadata XML — custom objects, custom fields, and more — directly inside the tool. The AI no longer has to write XML line by line, so metadata creation is faster and more accurate.

  • Supported metadata types: CustomObject, CustomField, ValidationRule, PermissionSet, GlobalValueSet, CustomLabel, RecordType, CustomTab, ListView

  • Batch creation: build multiple pieces of metadata in a single tool call, so even complex data models go up in one shot

  • Live preview: watch parameters being assembled by the Force tool in real time and confirm what’s being generated as it happens

  • Right-panel integration: clicking a Force tool call in a message expands the details so you can quickly inspect the output

  • Stronger validation: missing Picklist/MultiselectPicklist values, RichTextArea handling, automatic precision completion for formula fields — common pitfalls are detected and fixed automatically


2. Queue messages

You can now send the next message while Blaze is still running.

  • Auto-send: when the AI finishes its current task, queued messages are sent as the next instruction automatically

  • Queue display: a queue card above the input shows the messages waiting their turn

  • Instant dequeue: the Esc key or the stop button clears the queue immediately


3. Parallel tool execution in Flex mode

Flex mode now runs read-only tools (file viewing, SOQL queries, and similar side-effect-free operations) in parallel. Investigating multiple files or firing several queries at once now responds noticeably faster.

  • Safe parallelism: only side-effect-free tools are auto-detected and run in parallel, so data integrity is preserved

  • Parallel tool calls on the LLM side: the LLM request itself is set up for parallel tool calls so the AI can investigate multiple things simultaneously


4. Bug fixes

  • Skill command long-paste: fixed an issue where pasting a long block when finalizing a skill command hid the lower lines

  • Tool result ordering: fixed an issue where tool results appeared above the user message

  • Pinned conversation initial render: fixed an issue where older pinned conversations didn’t appear right after launching the app

  • Stop button responsiveness: removed a UI delay when pressing Stop so the queue stops immediately

  • Security hardening: strengthened session ID validation in auth callbacks and fixed multiple vulnerabilities