Turn Codex CLI runs into work your team can follow.

Start from a real brief instead of a prompt. Watch the run live, approve the commands it asks for, and inspect the diff before anything touches your branch.

Choose your download Watch it work
  • CLI login or your own API key
  • Scoped command approvals
  • Live logs, tests and token counters
2 ways
to execute: CLI or BYOK
Per-run
branch and worktree
Backend
enforced command policy

CODEX, SUPERVISED

It asks before it runs. You decide the scope.

Follow a Codex run from live logs to a command permission request, then to a file-by-file diff review and an operational dashboard.

01 / 05

Use the right agent for every job.

Bring your preferred coding agents into one operating layer and route work by column, role or tag.

  • Six supported agent families
  • Per-task model choice
  • Board-level routing rules

Showing Payments: Use the right agent for every job.

CodexClaude CodeGrokOpenCodeOpenAI-compatible

From individual Codex runs to an accountable workflow

Codex can carry substantial engineering work. What teams still need to decide is what should run, which context applies, which commands are allowed, and when the result is safe to accept. Those decisions should not disappear into terminal scrollback.

cadre.rocks gives Codex work a durable place on the board. The task explains the outcome, the run reports its state and logs, the command policy is enforced in the backend, and the approval stays with the person responsible for shipping.

WHY IT BEATS A RAW TERMINAL

1

Structured Codex tasks

Runs start from a brief with acceptance criteria, tags and a repository — reproducible and reviewable.

  • Acceptance criteria tracked as the run progresses
  • Task history, runs and tests kept per card
  • Comments become context for the next run
2

CLI login or your own key

Use the authenticated Codex CLI, or save a key for that agent in that column and keep API billing.

  • Credentials stored in the OS keyring, never in config files
  • Scoped to one agent in one column
  • Any OpenAI-compatible endpoint through a local profile
3

Commands you actually approve

Anything outside the board allowlist stops and asks. Approve once, trust it for the board, or reject it.

  • Bash allowlist per board
  • Sensitive files blocked by default, even after approval
  • Denied commands return a structured error, not a silent failure

HOW IT RUNS

From a defined task to an approved result.

  1. 01

    Connect your Codex setup

    Let cadre.rocks detect the CLI, or configure your own provider credential.

  2. 02

    Choose where Codex works

    Assign the agent, model and reasoning level to the relevant board column.

  3. 03

    Run and supervise

    Follow live logs and tests, and answer when the run needs a decision or a command.

  4. 04

    Inspect the diff

    Review file by file, approve, and keep the merge decision in human hands.

MODELS & KEYS

Use the models you already pay for. No token markup.

Every agent, in every column, runs one of two ways: through the CLI login you already have, or through your own API key. Requests leave your machine straight for the provider—cadre.rocks never sits in the middle of them.

CLI

Use the CLI login

No new billing

Point a column at a CLI that is already installed and authenticated. The run reuses that session.

  • Works with the subscription you already pay for
  • cadre.rocks detects which CLIs exist on this machine
  • Model and reasoning effort still chosen per column
  • No key to paste, nothing new to store
BYOK

Bring your own key

Your provider, your bill

Prefer API billing, a self-hosted endpoint or a model the CLI does not expose? Save a key for that agent.

  • Stored in the OS keyring—never in config files, logs or browser storage
  • Scoped to one agent in one column, separate from Chat BYOK
  • Any OpenAI-compatible endpoint through a local profile
  • Remove the credential and the column falls back to the CLI

Tokens are billed by your provider. cadre.rocks charges for the orchestration layer—boards, parallel runs, approvals—never per token.

Board settings → Column agents. The credential is scoped to that agent in that column, and is separate from Chat BYOK.

Choose your download

Free local boards · Your CLI logins and keys stay on your machine

SUPPORTED AGENTS

Bring the agents you already use.

cadre.rocks detects the coding CLIs installed on your machine and routes each task to the one you picked for that column. Same agents, same logins—now with a board, a queue and approval gates around them.

CC

Claude Code

Anthropic

Anthropic's engineering agent in the terminal.

claudeCLI · BYOK
CO

Codex

OpenAI

OpenAI's agentic coding CLI.

codexCLI · BYOK
GR

Grok

xAI

xAI's assisted build CLI.

grokCLI · BYOK
OC

Opencode

anomalyco

Open-source, multi-provider coding agent.

opencodeCLI · BYOK
{ }

OpenAI Compatible

Local profile

Endpoint and model defined by a local OpenAI-compatible profile.

profileBYOK

Agents run locally through their own CLI or through your key. Adding a model never requires a new account with us.

COMMON QUESTIONS

Questions, answered.

Does cadre.rocks add a token markup?

No. Provider usage is billed by the account or key you already use. cadre.rocks never sits between your machine and the provider.

Are API keys stored in the browser?

No. Credentials configured by the desktop app go into the operating system keyring — Keychain, Credential Manager or Secret Service — never into config files, logs or browser storage.

What stops an agent from running a destructive command?

The Rust backend enforces the policy. Commands outside the board allowlist require an explicit approval, and sensitive paths such as .env, *.pem and .ssh/* stay blocked even when a command is approved.

Can Codex and Claude Code share one board?

Yes. Supported agents can be assigned to different columns, tags or tasks in the same workflow and tracked side by side.

KEEP READING

Related guides.

See all guides →

CLAUDE CODE WORKFLOW

Claude Code

Keep your Claude Code login. Add tasks, parallel runs, blockers and diff approval around it.

Read the guide

MULTI-AGENT CODING WORKFLOW

Multi-agent workflow

Run several agents at once with isolated worktrees, a visible queue and ordered subtasks.

Read the guide

HUMAN-IN-THE-LOOP

Human in the loop

The four gates: questions, plans, commands and diffs — each with its context attached.

Read the guide

Put your agents to work.
Keep the final say.

Download cadre.rocks and move from scattered agent chats to tasks you can direct, inspect and approve.

Choose your download
Windows x64 macOS Universal Linux AppImage Linux .deb

Free local boards · No credit card · Windows, macOS & Linux