Sandbox: built end-to-end
Sandbox went from early access to a real, stable feature this cycle:- Sandboxes deploy fast — without a long wait to get one running.
- One-click WP Admin access. Links Foenix generates while working on your site (installing a plugin, opening a settings page) now log you in and take you straight there — no more dead links.
- Real-time deploy status. Sandboxes report Deploying → Live as they come up, and you can now launch multiple sandboxes in parallel.
- Transparent, correct pricing. 5GB (12 credits) / 10GB (20 credits) / 25GB (30 credits), with disk sizes that actually match what you picked.
- Free / Starter / Pro sandbox limits now enforced correctly across the board.
Agent system: fully reworked
- Three model modes. Lite (MiniMax M3), Standard (Gemini 3.5 Flash), Ultra (Claude Sonnet 4.6) — chosen per session, plus an independent image-model selector.
- Mode choice locks in per session once a task starts, so it can’t silently change mid-run.
- Orchestrator and Automation agents fully aligned — scheduled/automated runs now behave exactly like the agents you configure manually.
- Stopping a task no longer loses context for follow-up requests in the same session. Added a proper Stop button.
- Prompt-caching rework that meaningfully cuts wasted token cost on longer sessions.
- New Prompt Library and quick-start chips — one click to launch ready-made recipes (daily blog post with images, weekly SEO audit, and more).
- Faster responses in Lite mode via priority processing for MiniMax M3 and other lightweight models.
- Fixed token/cost accounting across our models — including a caching double-count bug and garbled Flash responses.
- Fixed autonomous agent lists not filtering correctly by site.
New
- Review and Publish. Choose where the result goes: Review prepares files for you to check first, Publish applies changes to the live site at once.
- Redesigned auth pages — sign-up, login, password reset, and confirmation.
- Dashboard tiles: new sites now appear at the top.
- File previews now render content directly in a modal (with a download button) for more file types.
- Renamed Vibe Coding → Live — opens a preview of your site where you can turn on Visual Edit mode.
- Universal links. Out of credits mid-task? Agents can now point you straight to the pricing page or the buy-credits modal.
🛠 Fixes
- Auth: Google sign-in occasionally failing on first attempt — fixed. Free-plan bonus credits now correctly granted on Google sign-up.
- UI: Long names now display correctly in your profile. Dashboard load-time fixes. General polish across profile, sessions, and mobile layouts.
- Plugin: Connection errors now shown clearly on the plugin settings page.
New
- Gemini 3.1 Pro now available on production across all agents.
- Credit packs now available on Starter plan. Previously limited to Pro only.
- Session titles now generated from chat context for easier browser history navigation.
🛠 Fixes
- Critical: WP plugin returning empty response on code execution — fixed, further improvements in progress. ⚙️
- Critical: Plugin not loading dashboard on first connect — fixed.
- Billing: Complete rewrite of credit counting logic.
- Billing: Overage no longer deducted from future purchases or renewals.
- Billing: Credit pack expiry corrected to 6 months.
- Billing: Upgrade/Downgrade blocked correctly when renewal not allowed.
- UI: Interface locked during backend requests — fixed, monitoring ongoing. ⚙️
- UI: Skeleton animation added to profile — prevents errors from clicking before data loads.
- UI: Edit agent panel moved to bottom of session with next run time and status visible.
- UI: Vibe Coding top bar layout fixed.
New
- Gemini 3 Flash added as a model option alongside Claude Sonnet 4.6. Choose per task — Flash for speed and cost, Sonnet for complex builds. Configure in Settings → Personality → AI Pipeline Models.
- Create Autonomous agents from any session. While in Conversational mode, just describe the task and tell Foenix when and how often to repeat it — for example, “every Monday at 9am, audit my SEO and email me the report”. Foenix expands it into a full prompt and shows a Schedule Task card to confirm.
- Stop a running task at any time. You can now interrupt a task before it finishes — useful when you spot it heading the wrong way or just want to cancel and try a different prompt.
- Per-agent model selection. In your Profile, you can now choose which AI model each agent uses individually — pick a lighter, cheaper model for simple recurring tasks and a more powerful one for complex builds.
🛠 Improvements
- A batch of bug fixes across the platform. We’re watching closely and shipping improvements continuously — keep the feedback coming.
Foenix.ai is in active early development. We’ve been building for the past 6 months and this is our current feature set.
What’s available today
- Multi-Agent architecture — Foenix operates as four specialised agents (Orchestrator, Architect, Coder, Verifier) with built-in retry-on-failure logic. See Architecture.
- Rollback — undo file-level changes (in active development)
- Safety modes — three levels (Safe, Balance, Hard) to control how strictly Foenix validates actions
- Multi-site dashboard — manage multiple WordPress sites from one account