Omnigent Files

Major new features

• 🔗 Connect a personal Databricks workspace: managed sandboxes mint per-user OAuth tokens on demand (server-refreshed), route the built-in jcode ACP harness and native coding harnesses through the AI Gateway, and can authenticate as an OAuth M2M service principal — no shared or static token required. (#6382, [#6916], [#6999], [#7401])
• 🪟 claude-sdk agents now show a context-window usage ring and support /compact for native compaction, including on a Claude Pro/Max subscription login. (#6119)
• 🎛️ Redesigned the new-session composer so workspace, model, effort, permissions, and folder browsing live in one place — plus fuzzy session search (⌘⌥S), a keyboard model picker (⌘⇧M), and a compact background-task badge. (#6345, [#7226], [#7413], [#7061])
• 🔀 Track and switch between multiple pull requests across repositories in a session, with automatic capture and manual link/unlink controls. (#6935)
• 🗄️ More ways to run Omnigent: CockroachDB as a supported database backend, Gensee as a built-in managed sandbox provider, Kubernetes node-pool tolerations for sandbox Pods, and starting a sandbox from multiple GitHub repos cloned in parallel. (#6360, [#7229], [#6881], [#6995])

Bug fixes

• 🔁 Sessions survive server and runner restarts: pending approvals and MCP elicitations stay answerable (even past the old few-minute cap), native Plans stay visible without a migration, and the Working indicator isn't dropped while a runner is still mid-turn. (#6696, [#6964], [#6629], [#6957])
• ↩️ Native Codex resume is far more reliable — cold resumes rebuild from the server transcript, permission overrides and picked reasoning effort are honored on resume, and recovered transcript items are replayed. (#7055, [#7147], [#7233], [#6860])
• ⏳ Chat status stays honest: no more stuck "Working…" after the agent finishes, and a completed Claude answer no longer collapses into the "Worked for" fold when a background task wakes the turn. (#6879, [#7002], [#7191])
• 🎨 Model pickers and the composer show readable catalog model names instead of raw provider IDs, and native model rate-limit errors are now retryable straight from chat. (#7260, [#7235])

• ⎋ Escape in the message input interrupts a working agent even when the client isn't streaming, and archived sessions stay archived instead of popping back into the sidebar. (#6871, [#6481])

• Agent uploads validate nested settings consistently, and tool approval retries preserve reviewed arguments. (#7457)

💜 Thanks to our community

This release was shaped by the people who filed issues, opened PRs, and talked through feature requests with us on our Discord! Thank you for building omnigent with us, keep the bug reports, ideas and contributions coming :)

Full Changelog: https://github.com/omnigent-ai/omnigent/blob/main/CHANGELOG.md