When NEXO first expanded beyond Claude Code, the brain became shared at the runtime layer before it became shared at the behavioral layer. Claude Code, Codex, and Claude Desktop could point at the same local memory system, but the lived experience still had two holes: Codex startup identity was inconsistent, and Deep Sleep still treated raw transcripts like a Claude-only source.
Version 2.6.14 closes those two gaps. That makes “one shared brain” much more honest as a product claim, because the system now aligns runtime wiring, startup behavior, and overnight analysis across both Claude Code and Codex.
Managed Bootstrap Files for Claude and Codex
NEXO now manages both bootstrap documents explicitly:
~/.claude/CLAUDE.md~/.codex/AGENTS.md
Both are now on the same migration contract and both use an explicit CORE / USER layout.
COREis owned by NEXO and can be updated safely as the product evolves.USERis owned by the operator and is preserved exactly across updates.
This matters because bootstrap files are no longer a fragile one-shot scaffold. They become a durable interface between product rules and personal instructions.
Codex Now Starts as NEXO, Not Just Codex + MCP
Adding the MCP server was never enough on its own. A user could open Codex, see the NEXO tools, and still get responses that felt like plain Codex with a memory plugin attached. That is not the product.
The fix in 2.6.14 is explicit: nexo chat and Codex headless automation now inject the current Codex bootstrap directly at startup. That means the session starts with NEXO identity and NEXO operating expectations, instead of relying on inconsistent global startup behavior.
Claude Code is still the recommended path, because its hook surface is deeper and its automation path is more battle-tested. But Codex is now much closer to product parity in how it actually starts.
Deep Sleep Now Reads Codex Sessions Too
The second gap was subtler but just as important. Deep Sleep could already run under a configurable backend, but its raw transcript collector still assumed Claude transcript storage. That meant a user spending the day in Codex could still get overnight analysis that was partially blind.
That is now fixed. Deep Sleep collection reads:
- Claude Code transcript JSONL files
- Codex durable session files and archived session files
The pipeline now keeps per-session client/source metadata and a stable mapping between logical session IDs and extracted transcript files. So mixed-client days no longer confuse the overnight phases.
Doctor Now Checks the Real Parity Surface
Runtime doctor checks now audit two things that used to be implicit:
- Bootstrap parity across the relevant clients
- Transcript-source parity for Deep Sleep inputs
That is important because it moves parity from “we think this should work” to “the runtime can tell you whether it actually does.”
Protected Bootstrap Files in Self-Modification Flows
NEXO's own internal protection layer now treats AGENTS.md as a protected file alongside CLAUDE.md. Evolution, watchdog, and self-audit flows are now aware that Codex bootstrap is part of the trusted control surface, not just a disposable side file.
What This Release Actually Means
- Shared brain is more real now, not just more marketable.
- Codex sessions can start with NEXO identity instead of generic Codex behavior.
- Operator-specific bootstrap customizations survive updates cleanly.
- Deep Sleep overnight analysis stays transcript-aware even on Codex-heavy days.
- Claude Code remains the recommended default, but Codex support is now materially stronger.
Install or Update
New install:
npx nexo-brainExisting installation:
nexo updateNeed to reapply shared-brain wiring and bootstrap files later?
nexo clients syncRead the full release history in the changelog or contribute feedback and fixes on GitHub.