Skip to content

Changelog

What shipped, and when.

  1. New plans and annual billing

    Pro and Business now include more, and both can be billed annually for two months free.

    • NewAnnual billing for Pro ($790 per year) and Business ($14,990 per year): ten months for the price of twelve.
    • ImprovedPro now includes 1 MCP Gateway server and 100 MCP Copilot requests a month.
    • ImprovedBusiness now includes 5 MCP Gateway servers, 500 MCP Copilot requests a month and one year of audit log retention.
    • NewSwitch plan or billing interval from the Billing page. Agent identities above your plan move with it.
    • ImprovedEnterprise is sized to your contract.
  2. MCP Gateway

    Put BLAZLE in front of your own MCP servers and decide, tool by tool, what your agents may run.

    • NewConnect your own MCP servers and point your agents at BLAZLE instead. Every tool call is checked by your rules, counted, and recorded against the token that made it.
    • NewSet each tool to Allow, Approve, Four-eyes or Block. Calls that need approval wait in Copilot Approvals and run once approved; blocked tools are hidden from the agent.
    • NewOne switch cuts every agent off from a server immediately, without touching the server or the agents.
    • NewServer credentials are encrypted and never shown again. What your servers return is checked before your agent reads it.
    • NewAgent Sweep now sends webhook events when an owner leaves, a credential is found in a repository, or an agent is quarantined or restored.
    • ImprovedThe billing page shows how many agent identities you use against your plan, and any identities above it.
  3. Agent Sweep

    See every agent that acts for your team, who owns it, and contain the ones nobody looks after.

    • NewAgent Sweep lists every API key and MCP token with its owner, last use and expiry, and flags the ones whose owner left, that were committed to a repository, that can change things but only read, or that sit unused.
    • NewQuarantine an agent in one click and restore it just as fast. Revocation goes through Copilot Approvals and needs a second person.
    • NewMCP tokens now stop working as soon as their owner leaves the workspace or is demoted to the token's level. Assign a new owner to bring one back.
    • NewWhen a member leaves, key managers are told which agents they left behind. On Business and Enterprise their API keys can be quarantined automatically.
    • NewThe GitHub App ties any BLAZLE credential committed in a pull request to the exact key or token. On Business and Enterprise, a repository scan also lists MCP configurations and agent instruction files.
    • NewThree new MCP Copilot tools: list_orphan_agents, quarantine_agent and revoke_mcp_token.
    • ImprovedThe dashboard and site are now available in Korean, Simplified Chinese, Japanese and Spanish, and your language follows your account to every device.
  4. Attack detection in every language, and on the way back

    BLAZLE now catches attempts to hijack your AI in many languages, and checks what flows back to your agents.

    • ImprovedThe AI check now understands attacks written in Vietnamese, Chinese, Indonesian, Thai and many other languages, including encoded and hidden-character tricks, and no longer flags messages that only talk about attacks.
    • NewPOST /intercept/output checks model answers and tool, web, file or email results before your agent reads them.
    • NewMCP tool results and resources are checked before they reach your agent: secrets are masked and suspicious data is marked as untrusted.
    • NewThe local proxy blocks destructive commands the model asks your agent to run, and uses the AI check for prompts.
    • NewThe GitHub App checks the files your coding agents follow (AGENTS.md, .cursorrules and similar) for hidden characters and planted instructions.
    • ImprovedThe proxy now follows your workspace's Standard or Strict mode, exactly like the cloud.
  5. Policy Library

    Bring your own security policies and enforce them in plain language.

    • NewUpload policies as PDF, Word, Markdown or text and ask them questions in plain language, including Vietnamese.
    • NewTurn any policy section into a rule from a few example messages. Rules warn first and block when you switch them on.
    • NewBlocked requests show the policy sections they relate to, in one click.
  6. AI detection in 50-100 ms

    The AI prompt-injection check now decides in 50-100 ms in production.

    • ImprovedMeasured across multiple production workspaces under attack traffic, down from 130-150 ms.
  7. Real-time notifications

    The dashboard now tells the right people the moment something needs them.

    • NewLive alerts for pending Copilot approvals, High and Critical intercepts, blocked pull requests, quota at 80% and 100%, and webhook delivery failures. Each alert goes only to members whose role can act on it.
    • ImprovedIDE intercept history with full totals and paging through older records.
    • ImprovedSecrets found in pull-request scans are masked in every stored record.
  8. Faster AI detection

    The AI prompt-injection check now decides in 130-150 ms in production.

    • ImprovedThe AI check runs only when it is needed, so most requests are decided in under a millisecond.
  9. Stronger API protection

    Safer webhooks and API traffic.

    • ImprovedStronger protection for webhook destinations and API traffic.
  10. Two-factor authentication

    Protect every account with TOTP.

    • NewTwo-factor authentication with any authenticator app, plus single-use backup codes.
    • ImprovedDashboard browsing no longer counts toward your request quota.
    • ImprovedGitHub App scans are metered without ever blocking a webhook.