NEXO started as a memory layer around Claude Code. That was the right first move, because Claude Code had the deepest hooks and the most battle-tested headless automation path. But it also created a product limit: if users wanted to live in Codex or share the same memory with Claude Desktop, the brain was shared only partially and the operational story was still too Claude-centric.
Version 2.6.12 fixes that product model. Version 2.6.13 then closes an important runtime confidence gap discovered immediately after release: a false-critical scenario around legacy personal KeepAlive helpers.
A Shared Brain Is Now the Baseline
The key product decision is simple: shared brain is not optional anymore. NEXO is the brain. What users choose now is everything around it:
- Which interactive clients they want connected
- Which client
nexo chatshould open by default - Whether they want autonomous background automation enabled
- Which backend should execute that automation
That means Claude Code, Codex, and Claude Desktop can now point at the same local NEXO runtime and memory system instead of behaving like disconnected islands.
nexo chat No Longer Assumes Claude Code
Before this release, nexo chat effectively meant “open Claude Code.” That was fine when NEXO only had one deeply supported terminal path. It became wrong the moment users started wanting Codex as their day-to-day terminal.
Now nexo chat opens the terminal client the user configured during install. The runtime also keeps model profiles per client/backend, so the selected provider is not just cosmetic.
Claude Code Is Still the Recommended Path
This is important and should be communicated clearly: Claude Code remains the recommended experience. It still has the deepest hook integration, the most hardened headless execution path, and the most mature operational surface in NEXO.
Codex is now supported in two important ways:
- As an interactive terminal client for
nexo chat - As a configurable automation backend for background jobs
Current recommended model profiles are:
- Claude Code: Opus 4.6 with 1M context
- Codex: GPT-5.4 xhigh
That said, Claude Code still has the richer native hook story, so the public-facing recommendation remains the same: if you want the most complete NEXO experience today, start with Claude Code.
Background Jobs Now Use a Shared Automation Runner
This is the architectural change that matters most under the surface. Core agentic jobs no longer need to inline provider-specific calls everywhere. They now route through a shared automation runner that reads the configured backend and model profile.
That means processes like synthesis, sleep, self-audit, postmortem, deep-sleep phases, immune triage, catch-up assessment, and watchdog repair can all execute under the selected backend instead of being permanently tied to Claude Code.
2.6.13: Better Self-Healing in Real Installations
Immediately after 2.6.12, a real-user runtime exposed an annoying gap: nexo doctor --tier runtime --fix could still leave a critical state even when the system was functionally okay. The culprit was a legacy personal daemon, wake-recovery, which lived as a manual KeepAlive helper outside the official personal-schedule model.
Version 2.6.13 fixes this the right way. Instead of ignoring the case, NEXO now supports personal keep_alive schedules officially, including the new restart_daemon recovery policy. Legacy wake-recovery installs are backfilled with official inline metadata and adopted into the normal reconcile flow.
The result is simple: fewer false criticals, more honest diagnostics, and less frustration for users who expect “doctor fixed it” to actually mean fixed.
What This Means for Users
- You can now keep the same brain across Claude Code, Codex, and Claude Desktop.
- You can choose which terminal client
nexo chatopens. - You can choose which backend runs NEXO's autonomous background work.
- You get stronger runtime self-healing, especially around managed personal schedules.
- You still get the most mature experience on Claude Code.
What This Means for Contributors
NEXO is increasingly a product that needs real-world feedback from multiple ecosystems, not just one terminal. If you use Claude Code, Codex, Claude Desktop, OpenClaw, or another MCP client, this is a good moment to get involved:
- Open issues when a client flow feels asymmetric or misleading
- Submit PRs for docs, install UX, compatibility testing, and site clarity
- Help validate client/model combinations in real production workflows
The project is still opinionated, but it is being built in the open and is much stronger when the public-facing copy, operational reality, and developer experience all match.
Install or Update
New install:
npx nexo-brain
Existing installation:
nexo update
Want to re-sync clients after adding a new one later?
nexo clients sync
Read the full release stream in the changelog or join the discussion on GitHub.