lazyagent Coding-Agent Monitor
Monitor multiple coding agents, sessions, states, and usage from one terminal or desktop dashboard.
- Source repo
- illegalstudio/lazyagent
- Stars
- ★ 194
- Last updated
- 2d ago
- License
- MIT
- Primary language
- Go
- FA score
- 75/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer running Claude Code, Codex, Cursor, and other assistants concurrently can watch their sessions from one terminal dashboard.
- Not a fit if
- Windows desktop users: desktop packages ship only for macOS and Linux
- Teams unwilling to accept unsigned Arch packages or the TrustAll tradeoff
- Users needing search and compaction for all nine agents; only some are covered
- Source review
- 75/100 · Good
What does this agent do, and when should you use it?
lazyagent is a locally run monitor for Claude Code, Cursor, Codex, Grok CLI, Kilo, Kimi Code CLI, Amp, pi, and OpenCode. It exposes the collected view through a terminal UI, a macOS/Linux desktop application, and a Bearer-token-protected HTTP API. It observes existing agent workflows instead of requiring users to launch sessions through a proprietary service, and it requires neither a server nor an account. Beyond live status, it can list and resume sessions, search transcripts, compact bulky session data, prune obsolete chat files, and report documented rate-limit or billing windows. Signed outbound webhooks can publish session transitions to systems such as Slack, custom dashboards, or CI endpoints.
lazyagent reads local session and transcript data produced by supported coding agents, identifies projects and activity states, and presents the result in its TUI, desktop card dashboard, or HTTP API. lazyagent sessions lists sessions associated with the current directory and can reopen a selection, while lazyagent latest immediately resumes the newest local session; both support YOLO-mode resumes. lazyagent search maintains an incremental local index for transcript-file agents—Claude, Codex, pi, Amp, Grok, and Kimi—and returns highlighted matches. lazyagent compact truncates large tool outputs, thinking blocks, and embedded images in place while keeping supported sessions resumable. lazyagent prune selects chat files by age or missing project directory and supplies agent selection, dry-run previews, and per-project confirmation. lazyagent limits reports the documented usage windows for Claude Code, Codex, Grok, Kimi Code, and Cursor, while webhooks emit signed JSON when session state changes.
- A developer running Claude Code, Codex, Cursor, and other assistants concurrently can watch their sessions from one terminal dashboard.
- A macOS or Linux user who needs visibility away from the terminal can use the menu-bar, tray, or detached desktop dashboard.
- A developer with a large archive of agent transcripts can search past conversations and compact oversized tool outputs, thinking blocks, and images.
- An engineer moving among repositories can list sessions for the current directory and resume a selected or most-recent session.
- A user managing usage limits can inspect the supported hourly, weekly, or monthly windows for Claude Code, Codex, Grok, Kimi Code, and Cursor.
- A team maintaining Slack alerts, an internal dashboard, or a CI endpoint can consume the HTTP API and signed state-transition webhooks.
How do you install or deploy this agent?
For the macOS desktop application, run brew install --cask illegalstudio/tap/lazyagent; this installs Lazyagent.app and links the lazyagent command. For a CLI-only macOS or Linux setup, run brew install illegalstudio/tap/lazyagent-cli. Arch Linux users can add the documented Illegal Studio pacman repository and run sudo pacman -Syu lazyagent; its packages are unsigned and the documented configuration uses SigLevel = Optional TrustAll. Linux desktop packages for amd64 and arm64 are also published through GitHub Releases as DEB, RPM, Arch, and AppImage artifacts. With Go installed, go install github.com/illegalstudio/lazyagent@latest builds the TUI-only edition. To build from source, run git clone https://github.com/illegalstudio/lazyagent, enter the repository, and use make tui; the complete desktop build requires Node.js and platform libraries, followed by make install and make build.
How do you use this agent?
Run lazyagent for the all-agent terminal dashboard. Restrict monitoring with commands such as lazyagent --agent claude, lazyagent --agent grok, or lazyagent --agent kimi. Use lazyagent --gui for the desktop application and lazyagent --api for the Bearer-token-protected HTTP API; set or rotate its credential first with lazyagent passphrase. lazyagent --tui --gui --api runs all three interfaces together. Session and maintenance workflows include lazyagent sessions, lazyagent latest, lazyagent search "query", lazyagent compact, lazyagent prune --days N, and lazyagent limits. Scripts can consume JSON from lazyagent sessions --json and lazyagent limits --json.
What are this agent's strengths and limitations?
- One local application monitors nine named coding agents and offers TUI, desktop, and HTTP API interfaces.
- Its observational model preserves existing terminal, IDE, and desktop workflows instead of imposing a replacement launcher or hosted account.
- It combines monitoring with concrete session operations: resume, indexed transcript search, pruning, compaction, and usage reporting.
- Desktop packages include the CLI, while a separate CLI-only build serves users who do not need a GUI.
- Bearer-token API protection and signed webhook payloads provide documented integration paths for private dashboards and automation.
- Desktop distribution is documented only for macOS and Linux; the supplied material provides no Windows installation or compatibility evidence.
- Feature coverage differs by agent—for example, transcript search and compaction support only specified subsets of the nine monitored products.
- The Arch repository distributes unsigned packages and requires accepting the documented
Optional TrustAlltradeoff or installing checksum-verified release assets instead. - API and webhook adoption adds passphrase, network, and signing-secret management responsibilities.
- A complete source build needs Node.js and platform libraries, while the Go installation produces only the TUI edition.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| lazyagent Coding-Agent Monitor This agent | 75 · Good | CLIFree | ★ 194 | 2d ago | Go | Codex · Claude Code |
| AgentAcct | 61 · Some gaps | CLIFree | ★ 755 | 1mo ago | Python | Codex · Claude Code |
| Huginn | 61 · Some gaps | Self-hosted serviceFree | ★ 50k | 2d ago | Ruby | — |
| EpicStaff Agent Workflow Platform | 57 · Major gaps | Self-hosted service | ★ 359 | 3d ago | Python | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The evidence describes a local observer requiring no account or server, a Bearer-protected HTTP API, signed webhook payloads, and maintenance commands with interactive confirmation, dry-run previews, and default backups. Release Actions are commit-pinned, macOS artifacts are signed and notarized, and checksums are published. Deductions apply because no implementation is supplied to verify file-access scope, token storage, webhook fields, or runtime least privilege; the Arch repository explicitly uses Optional TrustAll for unsigned packages; and complete recovery procedures after deletion are not shown. The repository, copyright holder, and inspirations are attributed, but publisher identity and individual maintenance responsibility are not independently established.
The README, license, Go module, and release workflow are strongly consistent about the product name, platforms, build process, and distribution formats, supporting full self-consistency. Dependencies are versioned, build prerequisites are described, and releases run tests and artifact checks. Dependency availability is reduced because Wails is an alpha dependency, the dependency graph is substantial, and no fallback strategy is shown. The workflow has explicit validation failures and diagnostic messages, but ordinary runtime errors for malformed sessions, unavailable agents, or failed webhooks are not evidenced.
The material thoroughly addresses terminal, desktop, scripting/API, and mobile-client scenarios and clearly bounds the product as an observer rather than a replacement for coding agents. Supported agents, interfaces, maintenance functions, macOS/Linux platforms, two architectures, and GUI/CLI build requirements are well delineated. Trigger precision is deducted because command entry points and explicit YOLO flags are clear, but no implementation demonstrates how session discovery, state transitions, webhook triggers, or agent identification avoid false positives.
The README has clear information architecture and points to documentation organized around getting started, concepts, interfaces, usage, maintenance, and reference topics. Installation, builds, launch modes, and package variants are thoroughly documented, and the complete MIT license matches the metadata. Deductions reflect evolving names and packaging (the history alias and cask/formula change), plentiful examples without an evidenced FAQ, and useful but incomplete limitation disclosure. News and Releases provide an update path, but no formal changelog is included. Organization, social, sponsorship, and bot identities are visible, while role ownership, a security-reporting path, and verified publisher identity remain unclear.
The TUI, desktop cards, HTTP API, JSON output, highlighted search results, resume actions, and maintenance previews provide directly usable outputs. Monitoring nine agent types without replacing their workflows offers plausible marginal value, but comparative superiority and the claimed 80+ MiB savings lack measurements or methodology. Open-source local operation, several free clients, and an optional paid iOS app support a reasonable cost-benefit case, although installation dependencies, desktop build complexity, configuration effort, and unsigned-distribution risk are not quantified.
Installation, dependency, licensing, and release-security claims are traceable to concrete README, go.mod, LICENSE, and workflow details, which corroborate versioned dependencies, tests, checksums, notarization, and platform artifacts. Deductions apply because the supplied evidence contains no implementation or test files for monitoring, compaction, deletion, indexing, authentication, or webhook signing; claims such as full open-source coverage, agent compatibility, and storage savings therefore remain partly asserted. Operational facts and marketing language are generally distinguishable, but measurements, assumptions, and inference boundaries are not explicitly labeled.
- The Arch repository distributes unsigned packages and requires SigLevel = Optional TrustAll; users unwilling to trust the hosting GitHub account should prefer release assets and verify their checksums.
- The product reads and indexes agent transcripts and can emit outbound webhooks. Before enabling it, verify the collected fields, index location, filesystem permissions, API bind address, token storage, and webhook destination.
- prune deletes sessions and compact rewrites files in place. Despite stated previews, validation, and default backups, use dry-run mode and independently back up important transcripts first.
- The --yolo mode for sessions/latest resumes agents with reduced confirmation constraints; use it only after understanding the originating agent's permissions and project risk.
- This assessment uses only the supplied static files; the program and packages were not executed, and publisher identity was not verified.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Changelog * a2d87391f6fffafca9ccb13c9f52c646e90cf1fb fix(limits): fall back to Cursor when Grok failsView on GitHub ↗ -
Release notes excerpt
Changelog * 618d09cd1d3e472911e645b0a5585cc76b9058e4 Merge pull request #67 from illegalstudio/feat/xai-subscription * 1b0c7cf3ef867005b91f2773e3fc03d67d942f06 feat(limits): support unified xAI subscriptions * 898183f217cf7a613732831e41d5742d8b23982a fix(limits): read shared xAI usage through CursorView on GitHub ↗ -
Release notes excerpt
Changelog * d09cfc3570632ac55d35db71e08ca6ef1ca30675 Merge pull request #66 from illegalstudio/feat/pacman-package * 1a5c72e55d295e7e658a389cbb301a0778319192 feat(release): publish the Arch package to illegalstudio/pacmanView on GitHub ↗
FAQ
Does lazyagent require a cloud account or dedicated server?
Does it replace or launch my coding agents?
What safeguards exist for pruning and compaction?
prune includes an interactive agent picker, dry-run previews, and per-project selection at confirmation. The source also states that rewrites are validated and backed up by default, but users should still inspect the preview before deletion.How is the HTTP API authenticated?
lazyagent passphrase, then start the service with lazyagent --api.