What We Shipped: Geta.Team v2.1.8 — Clone Employee, Codex Auto-Processes System Messages, and a Regex Bug We Owe You

Share
What We Shipped: Geta.Team v2.1.8 — Clone Employee, Codex Auto-Processes System Messages, and a Regex Bug We Owe You

v2.1.8 ships one big new thing — Clone Employee — plus three fixes we caught while testing it. Quick run-through.

Clone Employee

You can now duplicate an employee from the dashboard kebab menu. Click Clone, give the copy a name, and you get a brand-new employee with the same skills, the same memory, the same MCP servers, the same tags, and the same configuration as the source. The original is left completely untouched.

Things you'd expect to be cloned are cloned: every config field on the employee (skills, language, role, job title, photo, voice settings, agent type, sender/receiver permissions, performance rating), all related tags and contacts and MCP server connections, and the entire on-disk folder including the agent's persistent memory database, generated images, scheduled tasks, and uploads.

Things that are not cloned, by design: connector assignments (Slack, Teams, Google, Office365, WhatsApp, Telegram, Discord), phone numbers, webhook URLs, and widget tokens. These are unique bindings you re-attach per clone — it would be confusing if two employees suddenly shared the same Slack handle. History is also fresh — no inherited conversations, voice call logs, or audit trails. The clone starts with a clean slate of activity.

The new employee's CLAUDE.md and identity-bound skills (email-sender, task-scheduler, report-generation) get auto-patched to swap the source name and email for the clone's. Counters reset to zero. The clone's container spawns fresh on its first message.

If anything goes wrong during the copy (disk full, permission error), the partially-created employee is rolled back cleanly — no orphan rows.

The Clone option appears only for custom employees you own (or are admin over). Catalog-installed and read-only employees stay untouched.

Codex employees now respond to email, WhatsApp, and Slack messages immediately

This was a quiet bug worth flagging. If you had a codex-backed employee receiving messages from email, WhatsApp, Slack, Teams, Telegram, Discord, voice calls, Outlook, or webhooks, those messages were silently buffering instead of being processed. The agent only caught up when a human typed something in the chat — meaning a message that arrived at 9 a.m. could sit there for hours until someone happened to engage.

Claude and OpenCode employees never had this problem; they processed inputs as they arrived. We only noticed when testing the new Clone feature with a codex employee.

The fix is shipped. Codex now wakes up and processes system messages instantly, exactly like every other channel. If you have any backlog of buffered messages from before this release, they'll flush automatically the next time you open that employee's chat. No data loss.

OpenAI direct + GPT-5.5 now actually works (we promise this time)

The v2.1.7 release notes promised that custom_llm employees pointed at https://api.openai.com/v1 would correctly route GPT-5.5 (and other reasoning models) to OpenAI's new endpoint. They didn't. The check that decided whether to use the modern OpenAI adapter was subtly broken and silently fell back to the legacy one for the canonical OpenAI URL.

It's fixed in v2.1.8, with a stronger detection path that handles every URL shape (with or without protocol, with ports, with trailing slashes, mixed case). The Clone route was hardened in the same pass — every cloned custom_llm employee now boots with the correct adapter regardless of what the source had.

If you've set up any direct-OpenAI custom_llm employee since v2.1.7, the simplest fix is to re-save its cli_type once via the UI. The corrected route writes the right adapter, and a Clear Session picks it up.

New employees appear at the top of the creator's dashboard

Small but annoying: when you created a new employee (cloned or fresh), it landed at the bottom of your dashboard if you'd ever reordered your team. Now the creator sees their new employee at the top — where you actually want it. Other users still see new arrivals at the bottom of their reordered lists, so nobody else's custom ordering is disturbed.

Removed: Clone-VIP

The old Wikipedia-based "clone a VIP" employee generator is gone. The entry point had already been quietly dropped from the Hire flow weeks ago, leaving roughly 1,000 lines of dead code we cleaned up in this release. The new Clone Employee covers the duplication use case in a much more useful way.

Deployment

Standard build and deploy. No database migration. Frontend bundle changed, so a hard refresh (Ctrl+Shift+R) on already-open dashboard tabs picks up the new version display and the Clone menu entry.

That's v2.1.8. Clone is the heavy lift; the codex fix is the satisfying one.

Want to test the most advanced AI employees? Try it here: https://Geta.Team

Read more