Happier
Control, review, and ship work from multiple coding agents across all your devices.
Per-dimension scores and reasoning
The evidence clearly explains device-side end-to-end encryption, direct agent-to-provider traffic, and the routing metadata visible to the relay, earning full credit for data-flow transparency. It also shows job-scoped CI permissions, commit-pinned Actions, trusted-ref and release-actor guards, signing-key checks, and per-action surface/confirmation configuration. Deductions apply because the product can control terminals, files, Git pushes, pull requests, remote machines, and session sharing, while the supplied files do not establish runtime least privilege or mandatory confirmation for every consequential action; the voice assistant may also answer permission requests. Sensitive-data handling is supported mainly by README claims and a test harness rather than encryption implementation, key lifecycle, or recovery details. Worktrees, side-by-side release channels, and release retries offer recovery options, but no comprehensive undo path for user actions is shown. Attribution to Happy, the non-affiliation statement, and MIT labeling are clear; unknown publisher identity is not treated as suspicious.
The README, script inventory, and release workflows are strongly consistent about platforms, supported agents, release channels, and delivery mechanics. Workflows contain extensive input validation, source-SHA binding, timeouts, and explicit error exits, supporting full scores for consistency and failure messages. Frozen-lockfile installation, pinned versions, recovery paths, and broad database/provider test commands adequately address dependencies. Full credit is withheld because operation still relies on numerous model providers, app stores, EAS, registries, and OS tools, without evidence for every outage mode or offline fallback. No tests were executed for this static review.
The material addresses newcomers, terminal users, remote dev boxes, self-hosters, mobile, desktop, browser, multi-agent coordination, review, and release scenarios. It provides platform-specific installation plus SSH, Docker, Proxmox, hosted, and self-hosted options, demonstrating strong environment fit. Commands, registered session actions, workflow choices, and environment/platform parameters are precise. Capability boundaries lose a point because the README acknowledges agent-specific differences and links a feature matrix, and distinguishes stable, preview, and dev channels, but the matrix and detailed compatibility constraints are not part of the supplied evidence.
The README has strong organization across installation, startup, features, security, self-hosting, downloads, channels, and contribution paths. It includes many examples, stable product/command naming, enforced canonical SemVer, distinct release channels, and a changelog path. Deductions apply because there is no substantive FAQ, known limitations are not catalogued beyond dev-channel instability and capability differences, and the MIT license is supported by metadata and labeling but the license text itself is absent. Issue, discussion, Discord, contribution, and release-admin controls provide maintenance paths, but no named maintainers or independently verified responsible publisher are supplied.
The evidence presents actionable cross-device sessions, approvals, diff review, terminal and Git access, worktrees, cross-agent subagents, handoff, and a unified action registry. These features provide directly usable development outputs and clear marginal value over a single local agent terminal. Cost-benefit is not full because, although the project is free/open source, reuses existing subscriptions and keys, and supports self-hosting, the evidence does not quantify hosting, model, voice-service, operational, security-review, or multi-component deployment costs.
Most major claims point to specific documentation areas, while release workflows bind versions, source SHAs, signatures, candidates, and release notes for traceability. The package script inventory supplies some cross-source support through broad test and provider matrices. Deductions apply because the linked documentation, encryption implementation, lockfile, test results, and complete release pipelines are not included, so core claims such as end-to-end encryption, self-hosted feature parity, and comprehensive agent support cannot be fully verified here. The README distinguishes visible metadata, direct provider traffic, and dev-channel risk, but several marketing and security assertions are stated as facts without explicitly separating repository-supported evidence from unverified claims.
- Do not rely solely on the README's end-to-end-encryption claim for sensitive code. Review the actual protocol, key generation and storage, device revocation, recovery, and relay-metadata exposure.
- The product can remotely drive agents, terminals, and Git, share sessions, and let voice automation answer permission requests. Validate default confirmation policies, per-surface authorization, audit logs, and undo procedures before deployment.
- The recommended curl and PowerShell pipeline installers execute network-delivered scripts. Pin and verify installer contents, signatures, hashes, and update origins, preferably using reproducible or centrally controlled artifacts.
- Stable, preview, and dev channels have compatibility constraints; dev may break and cannot rely on the hosted cloud. Keep CLI, app, daemon, and server on a deliberately selected compatible channel.
- The publisher is not verified by the enterprise registry. Independently establish the responsible maintainers, vulnerability-reporting path, release-key controls, and long-term update ownership before enterprise adoption.
What does this agent do, and when should you use it?
Happier is a control plane for coding agents, comprising desktop, mobile, and web clients, a CLI, a machine-side daemon, and either a hosted or self-hosted relay. It starts and manages Claude Code, Codex, OpenCode, Pi, and other agents on the laptop, VPS, or dev box where the code resides, while synchronizing sessions to every signed-in device. Users can run parallel sessions, isolate them with Git worktrees, spawn subagents across agent products, switch engines within a session, and hand live sessions to another machine. Its workflow also covers approvals, queued steering, line-level diff review, file editing, an embedded terminal, and Git operations through pull-request creation. Session content is encrypted on the device before reaching the relay, while each underlying agent communicates directly with its own model provider; the same feature set is available through the hosted and self-hosted deployment paths.
The Happier daemon starts and manages agent processes on the machines that hold the user's code, while the Happier relay synchronizes end-to-end encrypted sessions with the phone, desktop, and browser clients. A user selects a machine, folder, agent, and model in the app, or starts from a project directory with happier, happier codex, or happier opencode. Happier can expose already-running Claude Code, Codex, and OpenCode sessions on other devices, allowing remote messages and approval decisions without migrating the session. A session may use its own Git worktree, launch review, plan, or delegate runs on another agent, and consume the returned result; live Claude Code and OpenCode sessions can also be handed to another machine. Review and delivery operations include browsing diffs, attaching feedback to exact lines, editing files, opening a live terminal, and staging, committing, pushing, branching, and opening pull requests. For automation, happier mcp serve exposes the shared action registry, while commands such as happier session list, happier session send <id> "rerun the failing test", and happier session actions execute <id> session.spawn_new control sessions from the CLI.
- A developer implementing several features in parallel can give each agent session its own Git worktree instead of letting concurrent changes share one checkout.
- An engineer away from their desk can monitor a running session from a phone, answer an agent's question, approve a permission request, and send further instructions.
- A team using both Claude Code and Codex can ask one agent to launch the other for review or planning, or switch the active engine while preserving the broader session.
- A user working across a laptop, VPS, and dev box can register remote machines over SSH and hand a live Claude Code or OpenCode session—and optionally its working tree—to another host.
- An organization that wants control over session storage can operate its own Happier relay while retaining end-to-end encrypted synchronization between devices.
What are this agent's strengths and limitations?
- One action registry serves the desktop, mobile, web, CLI, voice, in-session agent, and external MCP interfaces, with per-surface choices about execution and approval.
- It supports several built-in coding agents and permits ACP-compatible agents to be added from settings; sessions can also delegate across agent products or change engines midstream.
- Agent processes stay on the machines containing the code, model requests go directly to each provider, and session content is encrypted before it reaches the relay.
- Git worktrees, line-level diff feedback, file editing, a live terminal, and Git and pull-request operations extend the workflow from execution through review and delivery.
- The hosted and self-hosted relay paths use the same applications, encryption, and feature set rather than presenting self-hosting as a reduced edition.
- Happier does not supply model access; adopters still need suitable provider subscriptions, API keys, or supported local models.
- Capabilities differ by agent, so adding an ACP agent or switching engines does not guarantee equivalent handoff, review, or integration behavior.
- The relay processes session IDs, timestamps, turn status, and usage metadata even though code, prompts, messages, and transcripts are end-to-end encrypted by default.
- Preview and dev channels may contain incomplete or breaking changes; dev requires matching dev versions of the CLI, app, daemon, and server and cannot rely on the stable hosted cloud or a hosted dev web app.
- The complete system introduces clients, a daemon, a relay, and underlying agent processes, creating a broader deployment and troubleshooting surface than running one agent directly.
How do you install or deploy this agent?
The recommended route is the desktop app from https://happier.dev/download for macOS, Windows, or Linux; it configures the CLI and daemon. For a terminal-only installation, run curl -fsSL https://happier.dev/install | bash on macOS or Linux, or iwr https://happier.dev/install.ps1 -useb | iex in Windows PowerShell. The CLI is also available with npm install -g @happier-dev/cli. To add a remote machine, run happier machine setup --ssh user@host; pair it without a browser on the remote host using happier auth pair-remote --ssh user@host, or run happier auth login --no-open --method web remotely and approve the printed URL in another browser. Sign-in is required on first use, and the agents continue to use the user's existing provider subscriptions, API keys, or supported local models. To self-host the relay, run happier relay host install --mode system, or install it remotely with happier relay host install --ssh user@host --mode system.
How do you use this agent?
In the app, choose New session and select the machine, project folder, agent, and model. From a project directory in a terminal, run happier for Claude Code, happier codex for Codex, or happier opencode for OpenCode; the resulting session appears on signed-in desktop, mobile, and web clients. You can then queue or edit pending messages, steer an active turn, answer permission requests, inspect diffs, and return line-specific comments to the same session or a new one. Select a Git worktree when creating a session if concurrent tasks need separate branches and checkouts, and launch review, plan, or delegate subagents when work should cross agent products. For external automation, start happier mcp serve, or operate existing sessions with commands such as happier session list and happier session send <id> "rerun the failing test".
How does this agent compare with similar options?
Happier grew out of the maintainers' earlier contributions to Happy and pursues a faster-moving, more collaborative direction while acknowledging that foundation. Compared with operating one coding agent solely in its native terminal, Happier adds synchronized cross-device sessions, multi-machine control, cross-agent subagents, a unified approval inbox, and integrated review and delivery surfaces; the underlying agents still contact their own model providers directly.