The first stable release with the redesigned terminal experience and OpenTelemetry support.

New

A redesigned CLI. Every screen, picker, --help page and error message has a new, consistent look. Pickers show a live "n of m selected"; errors say what went wrong and the exact command that fixes it.
failproofai config, rebuilt. One guided flow:
- choose your agents;
- connect to FailproofAI Cloud, hosted or your own with --url;
- turn policies on.
OpenTelemetry, per agent. Send an agent's sessions to FailproofAI Cloud over OpenTelemetry instead of transcript upload:
- failproofai otel enable claude|codex|copilot|all switches an agent over.
- failproofai otel disable … puts its settings back exactly as they were.
- failproofai otel status shows your organisation's OpenTelemetry switch, each agent, and the local relay.
- failproofai otel env prints the exporter settings for any other OpenTelemetry app.
- --no-content leaves prompt and tool text out.
- --local sends through the failproofai daemon, which keeps batches on disk and retries if Cloud is unreachable.
Codex enforces policies headless. With administrator rights, config installs Codex's hooks where Codex runs them without an interactive trust prompt, so codex exec is covered too. config --status warns when Codex hooks aren't trusted.

Switching a machine to another organisation is a clean break. Nothing recorded before the switch is sent to the new organisation. The old organisation's cloud policies stop within about 30 seconds, and the new organisation's policies apply without restarting the daemon.
A new machine backfills its last week of sessions on its first connect.
Reconnecting with a different key replaces every saved credential. Events the old key was refused for are resent with the new key.
The daemon keeps collecting when a settings change or backfill arrives while it's busy.
Claude Code's OpenTelemetry export is accepted. It used to be refused because of how the key was written.
otel enable refuses while your organisation has OpenTelemetry off. It changes nothing, so your sessions keep arriving through transcript upload.
The Cloud key is never written for a plain-http remote address.
A network outage no longer discards OpenTelemetry batches. They're sent when the connection returns.
Codex OpenTelemetry sends conversation logs only, one session per run instead of dozens of empty ones. A log_user_prompt = false you set is kept.

After otel enable, restart the agent:
- start new Claude Code sessions;
- for Codex, quit it and stop its background app-server (pkill -f "codex app-server").
- A session that was already running isn't sent over OpenTelemetry.
Your organisation's admin turns OpenTelemetry on in FailproofAI Cloud under Settings → OpenTelemetry.

Send OpenTelemetry to Cloud. Point any OTLP/HTTP exporter at your Cloud address with a key that can add events:
```
OTEL_EXPORTER_OTLP_ENDPOINT=https://<your cloud>
OTEL_EXPORTER_OTLP_HEADERS=Authorization=Bearer%20<key>
```
Claude Code, Codex, Gemini CLI and any agent following OpenTelemetry's GenAI conventions appear in Sessions like every other session.
Off by default, per organisation. Admins turn it on in Settings → OpenTelemetry, or from the "connect with OpenTelemetry" option on an empty organisation's setup panel.
Protected under load. OpenTelemetry ingest has its own capacity and answers "try again shortly" when busy, so a flood of telemetry never slows sign-in, policies or alerts.
No empty sessions. Runtime-only traces that carry no AI activity are accepted and ignored.
The connection status needs read access to events, so a key that can only send data can't read your organisation's telemetry summary.
Supported transports: HTTP with protobuf or JSON, optionally gzip. For gRPC, put an OpenTelemetry Collector in front.