# Olomni

> Olomni is the neutral control for AI coding agents: one phone and one desktop companion (Olo) for several vendors (Claude Code, Codex, Cursor, Gemini CLI), instead of one app per vendor. Olo asks your permission before irreversible actions (on the desktop or from your phone), checks on your machine whether the AI really finished what it said (a green, amber or red verdict), hands the work to another AI when one runs out of quota, and keeps a memory shared across agents. Olomni never runs actions itself: it only carries your decision back to the agent.

Status: free beta, no card required (updated 2026-10-09; current version 0.1.36). Main platform: macOS 12+ (Intel and Apple Silicon). Windows 10/11 is in beta and may run one version behind; Linux is planned. Phone approvals work on any iPhone or Android through a web app at olomni.com/remote (no app store).

Integration depth, stated plainly: Claude Code is the most complete and best-tested integration. Codex, Cursor and Gemini CLI connect (live status, approvals, handoff) with limits while they are tested further. VS Code (Copilot), Windsurf, OpenCode, Cline and any MCP-capable agent connect over MCP.

## Quick answers
- What is Olomni? A neutral control for AI coding agents: one phone, several vendors, plus an independent check of whether the AI really finished.
- How is it different from Claude Code Remote Control or Codex in the ChatGPT app? Those are each vendor's native control for its own tool. Olomni covers several vendors in one place and adds the verdict and the handoff between AIs. If you only use Claude Code, Remote Control may be enough. See https://www.olomni.com/en/guides/olo-vs-claude-code-remote-control
- Does it replace my AI agent? No. It connects to the agents you already use.
- What stays on my computer? Your code, the text of your prompts and the AI's full responses. Only status, short permission requests, numeric summaries and saved memories go to your account.
- How much does it cost? Free during the beta.

## Pages
- [Home (English)](https://www.olomni.com/en): what Olo does, use cases, comparison, privacy model, FAQ.
- [Inicio (español)](https://www.olomni.com/): same content in Spanish.
- [Olo for Claude Code](https://www.olomni.com/en/claude-code): live status, phone approvals with the real effect, continuing with another AI when Claude Code hits its usage limit.
- [Olo for Codex](https://www.olomni.com/en/codex): approving Codex permission requests from the phone, shared memory, handoff.
- [Olo for Cursor](https://www.olomni.com/en/cursor): Cursor agent status, approval of risky terminal commands, handoff.
- [Olo for Gemini CLI](https://www.olomni.com/en/gemini-cli): live status, approvals, handoff.
- [Changelog](https://www.olomni.com/en/changelog): every Olo release, newest first.
- [Guides](https://www.olomni.com/en/guides): short answers and step-by-step guides, including comparisons with the vendors' own tools.
- [Full content for LLMs](https://www.olomni.com/llms-full.txt): all pages above as plain Markdown.
- [Privacy](https://www.olomni.com/privacy)
- [Terms](https://www.olomni.com/terms)
- [Press](https://www.olomni.com/press)

## What it does
- Live presence: Olo changes state (sleeping, thinking, working, needs approval, done) with each agent session, on top of any window including full-screen apps. Several agents at once are shown as lanes; subagents appear as mini-Olos.
- Approvals: irreversible actions (force push, deploy, deleting data) wait for the user's yes, on desktop or phone. Moving money, trading stocks and changing access permissions are blocked by policy. Users can choose "don't ask again in this project" for exact actions.
- Agent handoff: when one AI finishes a task, Olo records where it left off; the next AI opened in the same project (e.g. Codex after Claude Code) receives it at session start. The handoff is stored only on the user's computer, with secrets redacted. Olo also observes the project itself (git branch, uncommitted file names, whether the last test/build/lint passed) and gives that to the next AI, labelled as observed data; it also stays on the user's computer. A local MCP server (olomni-local: handoff_get, handoff_note, handoff_thread) lets any connected AI read that state and leave decisions, dead ends and next steps for the next AI; notes that look like instructions, links or secrets are held for the user's review. From any finished task, "Continue with…" copies a ready handoff for another AI and brings its app to the front. Agents can ask the user a question (ask_user) that reaches their phone and desktop; the user taps an option or types. Approval cards show what would really happen (e.g. commits a force-push would erase), computed by Olo on the user's computer, not the agent's own summary. When Claude Code hits its usage limit, Olo offers to continue with another AI; from the phone the user can hand a session off to another AI, and terminal agents (Claude Code, Codex, Gemini CLI) can open in the project with the handoff as their first message.
- Verdict: when an AI says it is done, Olo judges the turn by what it observed on the user's machine, not by what the AI said: green only if a test passed after the last change, amber if nothing was tested, red if a test fails, a protected path was touched or the AI is looping on the same failure. The user writes a project contract (rules, protected paths, the commands that show it is ready); Olo only runs the commands the user wrote, from a closed list of command types (tests, build, lint).
- Try it now: right after pairing a phone, Olo sends a harmless pretend permission request so the user can approve it with one tap in about a minute, without any AI connected.
- Morning report (opt-in): a push at 8:00 local time summarising yesterday's work with the user's AIs.
- Shared memory: facts agents save (e.g. "this project uses pnpm") are readable by all connected agents; each fact starts unconfirmed and the user can confirm, edit or delete it.
- "Your day": a summary of what the user's AIs finished today, to copy for a standup or share.

## Integrations
- Most complete and best tested (live status via hooks, approvals, handoff, strict mode on Mac, one-click connect): Claude Code.
- Connected with limits (live status, approvals, handoff): Codex, Cursor, Gemini CLI.
- Via MCP (Model Context Protocol, Streamable HTTP): VS Code (GitHub Copilot), Windsurf, OpenCode, Cline, and any MCP-capable agent.

## Privacy model
- Stays on the user's computer: the text of prompts typed to the agent, code and file contents, full AI responses, the handoff between AIs.
- Sent to the user's Olomni account: each session's status (e.g. "Editing payment.ts"), a numbers-only summary when a task ends (files, commands, minutes), permission requests (a short description of the action, up to 300 characters) and answers, messages the user sends to an agent from the phone, and saved memories. Session status is kept 7 days and permission requests 30 days. Users can export or delete everything from their Profile.

## Install
Create a free account, open Devices, generate a single-use command and paste it in Terminal (Mac) or PowerShell (Windows, beta). Olo installs, pairs with the account and connects the detected agents. Updates are automatic and cryptographically signed.

## Contact
- ops@olomni.com

---

# Full content

## Olo for Claude Code (Anthropic)
URL: https://www.olomni.com/en/claude-code

Olo shows what Claude Code is doing, asks before anything irreversible (even on your phone) and, when it runs out of quota, another AI picks up where it left off.

### Live status
Every Claude Code session shows up in Olo: thinking, editing a file, waiting for your permission or done. Several sessions at once, each in its own lane.

### Approvals with the real effect
Before a git push --force, a reset --hard or a deletion, Claude Code waits for your yes. The card doesn’t trust the AI’s own summary: Olo checks on your computer what would be lost, such as how many commits the push would erase.

### When it hits the limit
If Claude Code stops because of its usage limit or an overloaded service, Olo saves where it left off and offers to continue with Codex, Gemini CLI or Cursor with the full context. On Mac it opens the other AI in the project folder with the handoff already typed.

### From your phone
At olomni.com/remote you see your sessions, message them, stop them, resume a paused conversation or start a new task in your projects.

### It asks instead of guessing
Claude Code can ask you something (say, “Stripe or PayPal?”) and you answer with one tap from Olo or your phone.

### Memory and handoff
What it learns (“this project uses pnpm”) your other AIs can read, and you confirm or delete it. On start it knows where the project was left: branch, changes and whether tests passed.

### Setup
1. Create your free account and, in Devices, copy your single-use command.
2. Paste it into Terminal. Olo installs, pairs and detects Claude Code.
3. Olo adds its hooks to ~/.claude/settings.json, registers its MCP servers with claude mcp add --scope user and leaves a short block in ~/.claude/CLAUDE.md. Disconnecting removes all of it.

### FAQ
**How do I approve Claude Code actions from my phone?**
Open olomni.com/remote, add it to your home screen and pair it with a code. When Claude Code asks for permission you get a notification and decide with one tap; high-risk actions take two.

**What happens when Claude Code hits its usage limit?**
Olo gets the notice that the session stopped and why, saves where the work was left and offers to continue with another AI. It never switches AIs unless you choose to.

**Does Olomni read my code or prompts?**
No. Your prompts, your code and Claude Code’s responses stay on your computer. Only each session’s status, permission requests and the memories you control go to your account.

## Olo for Codex (OpenAI)
URL: https://www.olomni.com/en/codex

Olo shows what Codex is doing, lets you answer its permission requests from anywhere and gives it your other AIs’ context so it never starts from zero.

### Live status
Every Codex session shows up in Olo while it works, next to your other AIs.

### Its permissions, on your phone
When Codex asks to run a command, Olo intercepts it: you approve or reject on your desktop or phone, with what would really happen in view.

### Picks up another AI’s work
If Claude Code runs out of quota or you just want to switch, “Continue with Codex” hands it the handoff: branch, changed files, whether tests passed and the other AI’s notes. On Mac, Olo opens Codex in the project folder with the handoff as its first message.

### Shared memory
What Codex learns, Claude Code, Cursor and Gemini CLI can read, and vice versa. You confirm, fix or delete each memory.

### It asks instead of guessing
Codex can ask you a question with options and you answer with one tap.

### Setup
1. Create your free account and copy your single-use command in Devices.
2. Paste it into Terminal or PowerShell. Olo detects Codex by itself.
3. Olo adds its hooks to ~/.codex/hooks.json, its MCP server to ~/.codex/config.toml (with a backup) and a short block to ~/.codex/AGENTS.md.

### FAQ
**Can I approve Codex permissions from my phone?**
Yes. Olo intercepts Codex’s permission requests and brings them to olomni.com/remote: you decide with one tap and Codex carries on.

**How does Codex pick up Claude Code’s work without losing context?**
Olo observes the project on your computer (branch, changes, last test) and keeps the previous AI’s notes. Codex gets them on start, labelled as observed data or unverified notes.

**Does Codex work with Olo on Windows?**
Yes, on Windows 10 and 11. Opening it automatically in a terminal with the handoff is Mac-only for now.

## Olo for Cursor (Anysphere)
URL: https://www.olomni.com/en/cursor

Olo shows what Cursor’s agent is doing, holds risky commands until you say yes and passes it what your other AIs already know.

### Live status
Olo follows Cursor’s agent: when it gets your request, which files it edits and when it finishes.

### Risky commands, with your yes
Sensitive terminal commands the agent runs (force pushes, deletions, deploys) wait for your approval in Olo or on your phone.

### Handoff with your other AIs
“Continue with Cursor” copies a ready-to-paste handoff (where the project was left and the previous AI’s notes) and brings Cursor to the front.

### Shared memory
What Cursor’s agent saves, Claude Code, Codex and Gemini CLI can read. You decide what gets confirmed.

### It asks instead of guessing
The agent can ask you a question and you answer with one tap from Olo or your phone.

### Setup
1. Create your free account and copy your single-use command in Devices.
2. Paste it into Terminal or PowerShell. Olo detects Cursor by itself.
3. Olo adds its hooks to ~/.cursor/hooks.json and its MCP server to ~/.cursor/mcp.json, keeping a backup of the original file.

### FAQ
**Does Olo work with Cursor’s agent?**
Yes: live status through hooks, approval of risky commands, shared memory and handoff. After connecting, open Cursor and ask it anything: Olo confirms the first signal arrived.

**Can I hand Cursor’s work over to Claude Code?**
Yes. From “Latest” in Olo choose “Continue with…” and the other AI gets where the project was left plus Cursor’s notes.

**Does Olo change my Cursor settings?**
It only adds its entries to hooks.json and mcp.json, keeps a copy of the original and removes them if you disconnect Cursor in Olo.

## Olo for Gemini CLI (Google)
URL: https://www.olomni.com/en/gemini-cli

Olo shows what Gemini CLI is doing, asks before anything irreversible and gives it your other AIs’ context.

### Live status
Olo follows every Gemini CLI session: when it thinks, which tool it uses and when it finishes.

### Approval before anything irreversible
Before using a sensitive tool, Gemini CLI waits for your yes in Olo or on your phone, with what would really happen in view.

### Picks up another AI’s work
If Claude Code runs out of quota, “Continue with Gemini CLI” hands it the handoff. On Mac, Olo opens gemini in the project folder with the handoff as its first message.

### Shared memory
What Gemini CLI learns your other AIs can read, and you confirm or delete it.

### It asks instead of guessing
Gemini CLI can ask you a question with options and you answer with one tap.

### Setup
1. Create your free account and copy your single-use command in Devices.
2. Paste it into Terminal or PowerShell. Olo detects Gemini CLI by itself.
3. Olo adds its hooks and MCP server to ~/.gemini/settings.json and a short block to ~/.gemini/GEMINI.md.

### FAQ
**Is Olo compatible with Google’s Gemini CLI?**
Yes. Gemini CLI connects with live status through hooks, approvals, shared memory and handoff between AIs, with limits for now: we have tested it less than Claude Code.

**Can I continue with Gemini CLI when Claude Code runs out of quota?**
Yes. Olo offers it right away and, on Mac, opens Gemini CLI in the project with the full context.

**What Gemini CLI data leaves my computer?**
Only each session’s status, permission requests and the memories you control. Your prompts, code and responses stay on your computer.

## How to approve Claude Code permissions from your phone
URL: https://www.olomni.com/en/guides/approve-claude-code-from-your-phone
Updated: 2026-10-09

**Short answer:** Install Olo on your computer, open olomni.com/remote on your phone, add it to your home screen and pair it with a code from your profile. When Claude Code asks permission for something risky you get a notification and decide with one tap (two if the risk is high). No app store app is needed.

### Steps
1. Create your free Olomni account and, in Devices, copy your single-use command.
2. Paste it into Terminal (Mac) or PowerShell (Windows). Olo installs and connects Claude Code by itself.
3. On your phone open olomni.com/remote (Safari on iPhone, Chrome on Android) and add it to your home screen to get notifications.
4. Type a code from your profile to pair the phone.
5. Start a task in Claude Code. When it asks permission, tap Approve or Reject.

### What you see before deciding
The card doesn’t repeat what the AI says: Olo computes the action’s real effect on your computer. Before a git push --force, for example, it shows how many commits it would erase. Each card carries its risk level and a countdown.

### What is always blocked
Moving money, trading stocks and changing access permissions are blocked by policy: they never even become a question.

### What goes to your account
A short description of the action (up to 300 characters) and your answer. Your code, the text of your prompts and the AI’s responses stay on your computer. Olomni executes nothing: it only carries your decision back to the agent.

### FAQ
**Do I need to install an app on my phone?**
No. It’s a web app: open olomni.com/remote and add it to your home screen, on iPhone and Android.

**Does it work with Codex, Cursor and Gemini CLI?**
Yes. Claude Code is the best-tested integration; Codex, Cursor and Gemini CLI connect with limits; other MCP agents work for status and memory.

**Can I approve without being at my computer?**
Yes, as long as the computer stays on with the agent running and the phone is paired.

## What to do when Claude Code hits its usage limit
URL: https://www.olomni.com/en/guides/claude-code-usage-limit
Updated: 2026-10-09

**Short answer:** When Claude Code stops at its usage limit, don’t lose the thread: Olo saves where the work was left (branch, changed files, whether tests passed) and offers to continue with Codex, Gemini CLI or Cursor with that context already written. You choose; Olo never switches AIs on its own.

### Steps
1. Install Olo and connect Claude Code (one command, no keys to copy).
2. Work as usual: Olo keeps noting where each task stands.
3. If Claude Code stops at its limit or on an overloaded service, Olo tells you and gives the reason.
4. Choose Continue with… and the AI you want.
5. The other AI receives the handoff on start and picks up where Claude Code left off.

### What the handoff keeps
The git branch, the names of files with uncommitted changes and whether the last test, build or lint passed. Olo observes it in your project and gives it to the next AI labelled as observed data. It all stays on your computer and keys and tokens are redacted.

### How you continue
On Mac, Olo opens the other AI (Codex, Gemini CLI) in the project folder with the handoff as its first message. From your phone you can also hand the session to another AI.

### What it doesn’t do
It doesn’t switch AIs unless you choose to and it executes nothing for you: it only prepares the context.

### FAQ
**Does it work if the service is overloaded rather than my limit?**
Yes: Olo gets the notice that the session stopped and why, whether it’s the usage limit or an overloaded service.

**Is the handoff uploaded to the cloud?**
No. The handoff lives only on your computer.

**Which AIs can I continue with?**
Codex, Gemini CLI or Cursor.

## How to share memory between Claude Code, Codex and Cursor
URL: https://www.olomni.com/en/guides/shared-memory-between-ai-agents
Updated: 2026-10-09

**Short answer:** Connect your agents to Olo: what one learns (“this project uses pnpm”) is saved as an unconfirmed memory and the others read it on start. You see every memory in Olo’s panel and can confirm, edit or delete it; nothing is saved behind your back.

### Steps
1. Connect your agents to Olo (one click per agent in the panel, or with the install command).
2. Work with any of them: when it learns something durable, the agent saves it as a memory.
3. Open Olo’s Memory tab and review what’s new.
4. Confirm what’s right, edit what’s off and delete what’s not needed.

### How it works
Agents talk to Olo over MCP, the open standard for AI tools. When they start a task they consult memory; when they learn something durable they save it. Each memory starts unconfirmed and every connected agent reads it.

### What is not saved
The agents’ instructions say never to save secrets or credentials. Your code and prompts are not part of memory.

### Where it lives and how to delete it
Memories are stored in your Olomni account. You can export or delete them from your Profile.

### FAQ
**Which agents share the memory?**
Claude Code, Codex, Cursor and Gemini CLI, and any agent that speaks MCP over HTTP.

**Can the agent save something wrong or made up?**
Every memory starts unconfirmed and you decide whether it stays: confirm it, edit it or delete it.

**Is it the same as the handoff between AIs?**
No. Memory keeps durable facts your whole account knows; the handoff keeps where a task was left and lives only on your computer.

## How to hand off a task from Claude Code to Codex without losing context
URL: https://www.olomni.com/en/guides/hand-off-a-task-from-claude-code-to-codex
Updated: 2026-10-09

**Short answer:** With Olo, when a task ends the AI notes where it was left; the next AI you open in that project (say Codex after Claude Code) receives that note on start, along with the branch, changed files and whether tests passed. It all lives on your computer, with keys and tokens redacted.

### Steps
1. Connect Claude Code and Codex to Olo.
2. Work with Claude Code. When it finishes, it notes where the task stands.
3. Open Codex in the same project folder, or choose Continue with… in Olo.
4. Codex receives the handoff as context on start and picks the work up.

### What the handoff includes
What the AI noted (decisions, dead ends, next step) and what Olo observes by itself: git branch, names of uncommitted files and whether the last test, build or lint passed. Observed data is labelled as such, to tell it apart from what the AI claims.

### What keeps it safe
Notes that look like instructions, links or secrets are held for your review instead of passing to the next AI. Keys and tokens are redacted and the handoff never leaves your computer.

### How the AIs use it
Olo offers a local MCP server (olomni-local) with three tools: handoff_get to read where the project was left, handoff_note to leave a decision, dead end or next step, and handoff_thread for the working thread.

### FAQ
**Does it work between any pair of AIs?**
Between Claude Code, Codex, Cursor and Gemini CLI, and any agent that speaks MCP.

**Are my notes uploaded to the cloud?**
No. The handoff is stored only on your computer.

**Do I have to copy and paste the context?**
No. With “Continue with…” Olo prepares the handoff and opens the other AI in the project folder.

## How to stop an AI agent from force-pushing or deleting data
URL: https://www.olomni.com/en/guides/stop-an-ai-agent-from-force-pushing
Updated: 2026-10-09

**Short answer:** Put a gate between the agent and irreversible actions. With Olo, a git push --force, a reset --hard, a deploy or a deletion waits for your approval, and the card shows the real effect computed on your computer, not the AI’s own summary. Moving money, trading stocks or changing access is blocked by policy.

### Why asking the AI to be careful isn’t enough
A prompt is not a control: the AI can be wrong or be fooled by content it reads. The protection has to live outside the model, in a gate the agent can’t skip.

### How the gate works
Olo hooks into the agent’s own hooks and into MCP. When the agent is about to do something irreversible, it waits for your answer on desktop or phone. The card shows what would really be lost, such as how many commits a forced push would erase, checked on your machine.

### Rules so you don’t get tired
You can choose “don’t ask again in this project” for exact actions, so repetitive things don’t interrupt you while risky ones still ask.

### What Olo doesn’t do
It doesn’t execute the action: it only carries your decision to the agent. Nor does it replace a sandbox or code review; it’s a layer of human decision over what’s irreversible.

### FAQ
**Which actions ask for approval?**
Irreversible ones: force push, reset --hard, deploys, data deletion. Moving money, trading stocks and changing access permissions are blocked.

**Does it work with Cursor, Codex and Gemini CLI?**
Yes, as well as Claude Code.

**Can I approve from my phone?**
Yes, at olomni.com/remote; high-risk actions take two taps.

## What is an AI agent approval layer?
URL: https://www.olomni.com/en/guides/what-is-an-ai-agent-approval-layer
Updated: 2026-10-09

**Short answer:** An approval layer is software that sits between an AI agent and actions with real consequences (deleting, publishing, spending) and requires an explicit human decision before they run. Olomni is one: it shows what the agent is doing, asks permission with the real effect and carries your decision back, without executing anything itself.

### What problem it solves
Coding agents already work alone for minutes or hours: they ask when you’re no longer looking and tomorrow they’ve forgotten the context. An approval layer gives visibility, control over what’s irreversible and continuity.

### What it needs
- Live visibility of what each session of each agent is doing.
- Approval with an independently verified effect, not the agent’s own summary.
- Access from anywhere, like the phone, because the agent doesn’t wait for you to get back to the desk.
- Blocking policies for what must never happen, like moving money.
- Memory and handoff so switching agents isn’t starting from scratch.

### What it isn’t
It isn’t a sandbox and doesn’t replace code review: it reduces the risk of irreversible actions by putting a person in the decision.

### How Olomni does it
Olomni works with Claude Code, Codex, Cursor and Gemini CLI through their hooks and MCP. Olo, a character on your screen, shows the status; your phone receives the approvals; and a shared memory you confirm saves you from repeating context.

### FAQ
**Is it the same as a sandbox?**
No. A sandbox limits what the agent can touch; an approval layer decides, with a person, what should actually happen.

**Does it slow the agent down?**
It only stops at what’s irreversible; everything else runs uninterrupted.

**Which agents does Olomni support?**
Claude Code in depth; Codex, Cursor and Gemini CLI with limits; VS Code (Copilot), Windsurf, OpenCode, Cline and any MCP-over-HTTP agent.

## Olo or Claude Code Remote Control: which to use and when
URL: https://www.olomni.com/en/guides/olo-vs-claude-code-remote-control
Updated: 2026-10-09

**Short answer:** If you only use Claude Code and following the session and approving from your phone is enough, Remote Control (Anthropic’s native feature) is probably enough. Olo adds three things: one control for several AI vendors (Claude Code, Codex, Cursor, Gemini CLI), an independent check of whether the AI really finished (green, amber or red based on what Olo sees on your machine), and handoff to another AI when you run out of quota.

### What Remote Control does
Remote Control is Anthropic’s feature for continuing a local Claude Code session from your phone or the browser, with notifications when it needs something from you. It is native: nothing extra to install. It asks you to sign in with a Claude subscription account, not an API key. See the official documentation for the current requirements.

### What Olo adds
- One control for several vendors. Claude Code, Codex, Cursor and Gemini CLI on the same phone and the same Olo, instead of one app per vendor. Claude Code is the best-tested integration; the others connect with limits.
- Checking “done”. When the AI says it finished, Olo looks at what it saw on your machine: green only if a test passed after the last change, amber if it wasn’t tested, red if it fails or touched a path it shouldn’t.
- Handoff between AIs. If one AI runs out of quota, the next one starts knowing where the project was left.
- Real effect before approving, and undo. The card computes on your computer what would really happen and saves a restore point.

### Where Remote Control wins
It is Anthropic’s own, needs nothing installed, and lets you carry the whole Claude Code session from your phone. Olo is an add-on: alerts and approvals, status, handoff and verdict.

### Olo’s limits today
Olo is in free beta. The main platform is Mac (Windows is in beta and may run one version behind). Claude Code is the best-tested integration; Codex, Cursor and Gemini CLI work with limits. The verdict is a new feature, from October 2026.

### FAQ
**Does Olo replace Remote Control?**
No. It is an add-on for people who use more than one AI vendor or want something to check the AI’s “done”. If you only use Claude Code, start with Remote Control.

**Does Olo need the Claude subscription that Remote Control asks for?**
Olo doesn’t depend on Remote Control: it connects through hooks and MCP. Claude Code itself still needs its own access.

**How much does Olo cost?**
It is free during the beta, no card needed.

## Controlling Claude Code and Codex from your phone: options in 2026
URL: https://www.olomni.com/en/guides/control-claude-code-and-codex-from-your-phone
Updated: 2026-10-09

**Short answer:** There are three routes today. Each vendor’s native option (Remote Control in Claude Code; Codex inside the ChatGPT app), third-party apps that cover Claude Code and Codex (such as the open-source Happy Coder, and Omnara), and Olo, which brings several vendors under one control and checks whether the AI really finished. Pick the native one if you use a single vendor; a neutral control if you use several. Reviewed on October 9, 2026: these tools change fast.

### The options
- Remote Control (Claude Code). Anthropic’s native feature for following and approving Claude Code sessions from your phone. Documentation.
- Codex in the ChatGPT app. OpenAI lets you continue and approve Codex tasks from your phone while it runs on your computer. OpenAI announcement.
- Happy Coder. Open-source mobile and web client for Claude Code and Codex. happy.engineering.
- Omnara. Phone app for Claude Code and Codex. App Store.
- Olo. One control for Claude Code (in depth), Codex, Cursor and Gemini CLI, with an independent check of whether the AI finished and handoff between AIs. Free beta, Mac.

### How to choose
- You use only Claude Code: start with Remote Control.
- You use only Codex: the ChatGPT app.
- You use Claude Code and Codex and want one app: Happy Coder or Omnara, or Olo if you also want to check “done” and pass work from one AI to another.
- You use Cursor or Gemini CLI alongside those: Olo is the option designed to bring them together, with limits.

### What to check before choosing
- Which vendors it really covers (not every integration is equally deep).
- What goes to the cloud: code, prompts or only status.
- Whether it requires a specific subscription account.
- Which systems it runs on (Mac, Windows, Linux).

### FAQ
**Can I approve Claude Code permissions from my phone without installing anything?**
Yes, with Remote Control, Claude Code’s native feature, if your account meets its requirements. See Anthropic’s documentation.

**Is there a free option?**
Remote Control and the ChatGPT app are part of their plans; Happy Coder is open source; Olo is free during the beta. Prices change: check each site.

**Which covers the most vendors?**
Olo connects Claude Code, Codex, Cursor and Gemini CLI (Claude Code in depth, the rest with limits). Happy Coder and Omnara cover Claude Code and Codex.

## How to tell if an AI coding agent really finished (not just said “done”)
URL: https://www.olomni.com/en/guides/how-to-tell-if-an-ai-coding-agent-really-finished
Updated: 2026-10-09

**Short answer:** Check whether a test (tests, build or lint) ran after the last change and passed. If it wasn’t tested, it isn’t finished even if the AI says “done”; if the test fails, it isn’t either. Olo automates this on your machine: green if a test passed after the last change, amber if it wasn’t tested, red if it fails or touched a path it shouldn’t.

### Steps
1. Check what really changed with git status and git diff, not with the AI’s summary.
2. Run the project’s tests and build after the last change (or ask the AI to and read the result).
3. Check it didn’t touch sensitive paths: migrations, configuration, deployment, key files.
4. Be wary if it repeats the same error several times: ask for another approach or hand the work to another AI.
5. With Olo, write once in the project contract which commands show it’s ready; Olo runs them on your machine and marks the turn green, amber or red.

### Why an AI’s “done” isn’t enough
An agent describes what it thinks it did. It may have edited files without running the tests, or run a test before its last change. The useful evidence is what was observed on your machine after the last change.

### What each color means in Olo
- Green: an observed test passed after the last change.
- Amber: nothing was tested after the last change.
- Red: the test fails, it touched an untouchable path, or the AI is looping on the same failure.

### What it doesn’t do
It doesn’t replace code review or guarantee there are no bugs: it only checks what your tests cover. Olo only runs the commands you wrote in the project contract.

### FAQ
**Does it work with any AI?**
The verdict is computed from what Olo observes on your machine, through each AI’s integration. Claude Code is the best-tested; the others work with limits.

**Does Olo run code from the AI?**
No. It only runs the commands you wrote in the contract, from a closed list of command types (tests, build, lint).

**What if I have no tests?**
The turn stays amber: no tests, no evidence. A good reason to add at least one.

## Changelog
- Olo 0.1.37 (2026-10-09): Olo gets a new look: three styles to choose from (Status, Porcelain and Graphite) in the panel and on your phone. In “Status” the background is what is happening: mint when calm, yellow when something waits for your answer, pink when something failed. The panel folds itself away when it is no longer needed and leaves just Olo, or with the arrow in its header. What waits for your OK shows in a strip outside “Now”, and the phone verdict reads without knowing how to code.
- Olo 0.1.36 (2026-10-09): “Try it now”: as soon as your phone is linked, Olo sends you a pretend permission request and you approve it with one tap, no AI needed. In a minute you see how the real, sensitive ones will reach you, and it tells you whether you decided from the phone or the computer.
- Olo 0.1.35 (2026-10-09): Olo judges each turn by what it saw on your machine, not by what the AI said: green only with a test observed after the last change, amber if it was not tested, red if it fails or touches a protected path. It warns you when an AI loops on the same failure, and introduces the project contract: your rules, protected paths and how "done" is checked, for any AI.
- Olo 0.1.34 (2026-10-05): Strict mode: your approvals move into the system configuration, so no AI can remove them or approve itself. Sensitive actions can only be approved from the phone you link with the code Olo shows you, verified by the server. Olo restores its hooks if someone removes them, and the checkup tells you how secure each AI is.
- Olo 0.1.33 (2026-10-05): Olo finds Codex even when it ships inside the ChatGPT app: it shows up on your phone to start tasks and to review another AI’s work.
- Olo 0.1.32 (2026-10-05): Your AIs review each other: from your phone, ask another AI to check what the first one did and get a verdict (it never sees your .env files or keys, and secrets are redacted). You also see what each AI spends and, when it reports it (Codex), how much quota is left. A cleaner, tidier phone app.
- Olo 0.1.31 (2026-10-04): Ask and review from your phone, no coding needed: tell Olo what you want and pick the AI (Claude Code, Codex or Gemini). Watch each task go from “Requested” to “Done”, see which files changed, ask for a tweak or undo it all. Approvals in plain words, and you choose how much freedom they get.
- Olo 0.1.30 (2026-10-04): Your next AI gets a more reliable project state: anchored to the repo root, with a summary of each repo, and never trusting a test whose result wasn’t seen. If Olo crashes it reopens itself, and it tells you about anything sensitive an AI did while Olo was closed.
- Olo 0.1.29 (2026-10-04): Olo opens at login so your approvals, memory and phone are never left without it, and only one copy can run at a time.
- Olo 0.1.28 (2026-10-04): Olo looks sharper and livelier: new lighting, a vinyl finish, a glass screen on its chest and smooth transitions between states. It uses less memory.
- Olo 0.1.27 (2026-09-27): Your AIs ask instead of guessing: answer with one tap from Olo or your phone. Before approving something sensitive you see what would really happen, checked by Olo on your computer. Hand work to another AI from your phone and, if Claude Code runs out of quota, Olo offers to continue with another one and opens it in the project folder. A simpler panel: what needs your answer shows at the top, covering nothing.
- Olo 0.1.26 (2026-09-27): Your AIs hand work to each other: Olo records where each project was left (branch, changes, whether tests passed) and tells the next AI. A “Continue with…” button to switch AIs in one tap, and your AIs can leave each other notes (decisions, dead ends); anything suspicious is held for your review. Panel redesigned in the olomni.com style.
- Olo 0.1.25 (2026-09-26): Lighter while working: Olo, its orb and the mini-Olos are drawn together (−28% CPU with an agent working). In low-power mode or when your computer is hot, effects slow down by themselves. Hiding Olo from the tray pauses it completely.
- Olo 0.1.24 (2026-09-26): Lighter at rest: a still pose with blinks, the orb at 20 fps and mini-Olos that wake up half as often. Security: high-risk actions can no longer be approved by dragging the card; you confirm with “Approve”.
- Olo 0.1.23 (2026-09-26): Olo looks at you and follows you with its head, charges an energy orb while your AI works and jumps to the window it worked in after a big task. Drag a card onto Olo to approve or fling it to reject, rewards when you level up, and “Your day” in 15 seconds.
- Olo 0.1.22 (2026-09-25): Security: 8-digit pairing codes. Olo keeps showing on top of full-screen apps.
- Olo 0.1.21 (2026-09-25): A fix in the approval card, and Olo stays visible over full-screen apps on Mac.
- Olo 0.1.20 (2026-09-25): Olo finally shows on top of full-screen apps on Mac.
- Olo 0.1.19 (2026-09-25): Handoff between AIs: each AI knows where the project was left. “Latest” with “Hand to another AI”, one-tap feedback and a lighter panel.
- Olo 0.1.18 (2026-09-25): Olo moves across the whole screen, full-screen apps included, and the panel has a new design.
- Olo 0.1.17 (2026-09-25): A guided first minute, friend invites (3 each) and the Friend bow.
- Olo 0.1.16 (2026-09-25): Olo updates itself, stores its credentials without the keychain, and the Founder laurel arrives for the first 100.
- Olo 0.1.15 (2026-09-24): Olo grows with you: it earns XP from what your AIs finish, levels up, keeps your streak and unlocks 10 accessories. Share your day with a card made on your device.
- Olo 0.1.14 (2026-09-24): Your phone, with Olo inside: start new tasks in your projects and ask permission to message your AIs. Each AI gets its own mini-Olo.
- Olo 0.1.13 (2026-09-24): Connect your phone by scanning a code, and your AIs update themselves.
- Olo 0.1.12 (2026-09-24): Messages from your phone always arrive: your AIs wait for you when done and you can resume a paused Claude Code conversation.
- Olo 0.1.11 (2026-09-24): Your phone as a remote: see what your AIs are doing, message them, stop them and give your OK from your phone.
- Olo 0.1.10 (2026-09-24): Mandatory approvals for sensitive actions, confirmed memory at the start of each session, diagnostics and 8 compatible AIs.
- Olo 0.1.9 (2026-09-23): Olo in 3D, colored mini-Olos, a tabbed panel and half the memory use.
- Olo 0.1.7 (2026-09-23): Mini-Olos: each agent working at the same time orbits Olo in its own color, with its subagents following in line.
- Olo 0.1.6 (2026-09-23): One Olo for all your agents: several at once, each in its own lane.
- Olo 0.1.5 (2026-09-23): Expressive Olo: 12 expressions, a summary when done, and your Olo with its own name, color, size and tone.
- Olo 0.1.4 (2026-09-23): macOS no longer blocks the app when installed with the command.
- Olo 0.1.3 (2026-09-23): One-command install: Olo pairs itself and connects your agents on first launch.
- Olo 0.1.2 (2026-09-23): Your agent notifies Olo by itself (hooks) in Claude Code, Codex, Cursor and Gemini CLI.
- Olo 0.1.1 (2026-09-23): The character is named Olo. Connect Claude Code, Codex, Cursor and Gemini CLI in one click.
- Olo 0.1.0 (2026-09-21): First beta.