Skip to content
Tekai

AgentManager — CLI/TUI for Managing AI Coding Agent Installations

Kevin Elliott April 11, 2026 open-source-project medium credibility
View source

Referenced in catalog

AgentManager

AgentManager — CLI/TUI for Managing AI Coding Agent Installations

Source: github.com/kevinelliott/agentmanager | Author: Kevin Elliott | Published: 2026-01-08 Category: open-source-project | Credibility: medium

Executive Summary

  • AgentManager (agentmgr) is a Go CLI/TUI that acts as a unified package manager for AI coding agent CLIs. It detects installed agents (Claude Code, Aider, Amp, Gemini CLI, OpenCode, GitHub Copilot CLI, Goose, and 25+ others), checks their versions against upstream registries (npm, PyPI, Homebrew, GitHub Releases), and enables install/update operations via a single command interface. It ships as v1.0.24, is MIT-licensed, and is installable via Homebrew or go install.
  • The problem it solves is real and growing: developers working with multiple AI coding agents routinely juggle npm, pip, brew, and native installers to keep tools current. As the AI coding agent ecosystem fragments into 30+ CLI tools, a unified management layer becomes genuinely useful.
  • The execution is functional but limited by solo authorship: 19 GitHub stars, 98/100 commits from a single developer (Kevin Elliott), and a velocity slowdown since February 2026 (last release v1.0.24, February 28, 2026 — no releases in 6 weeks at review time). The REST/gRPC APIs and plugin system are more ambitious than the project’s current community size can sustain.

Critical Analysis

Claim: Detects and manages 32+ AI coding agents

  • Evidence quality: verifiable (open-source catalog.json)
  • Assessment: The catalog.json file confirms 32 agents at the time of review, covering the major AI coding CLIs (Claude Code, Gemini CLI, Aider, Amp, OpenCode, GitHub Copilot CLI, Goose, etc.) and several niche tools (Tokscale, Droid, Dexter, Plandex). Detection logic probes npm, pip, Homebrew, native binary paths, and package registries. Version checking sources from npm registry, PyPI, Homebrew API, and GitHub Releases depending on the agent’s primary distribution channel.
  • Counter-argument: The catalog is only as current as the maintainer keeps it. The AI coding agent landscape changes weekly — agents launch, rename, and deprecate constantly. With a single maintainer and slowing commit activity, catalog freshness is the primary long-term risk. The agentmgr catalog refresh command fetches from the GitHub repo, which means the tool’s utility degrades if the author disengages.
  • References:

Claim: Cross-platform (macOS, Linux, Windows)

  • Evidence quality: verifiable (CI workflows)
  • Assessment: The CI configuration runs integration tests on Ubuntu, macOS, and Windows (added in v1.0.13). The Homebrew formula and go install paths are confirmed cross-platform distribution methods. However, an open issue at review time — “fix: add missing showNativeManageAgentsWindow stub for non-darwin platforms” — reveals that the systray/background helper remains macOS-first. The core CLI detection and management commands work cross-platform; the background update helper does not.
  • Counter-argument: For most use cases (CLI usage, not background notifications), the cross-platform claim holds. Teams on Linux or Windows CI can use agentmgr agent list --format json for scripting without hitting the darwin limitation.

Claim: REST and gRPC APIs for integration

  • Evidence quality: verifiable (codebase, agentmgr api docs)
  • Assessment: The REST API and gRPC interface exist and are accessible via agentmgr api docs / agentmgr api spec (OpenAPI). The API exposes catalog data, agent detection results, and installation operations. However, the README contains no documented integration examples, no authentication mechanism, no TLS configuration, and no versioning contract. These APIs appear to be developer-convenience features rather than production-ready integration surfaces.
  • Counter-argument: The Go library (pkg/detector, pkg/catalog, pkg/installer) is the more appropriate integration path for teams wanting to embed AgentManager logic. The library has clean package boundaries and can be used without the security ambiguity of an unauthenticated HTTP server.
  • References:

Claim: Detection plugin system for custom agents

  • Evidence quality: verifiable (docs/plugins.md, agentmgr plugin command group)
  • Assessment: The plugin system is real and documented. Teams can define custom agent detection logic in YAML/JSON and register it via agentmgr plugin create. This is a meaningful differentiator — it allows organisations with internal AI tooling or unlisted agents to extend the catalog without forking the project. Added in v1.0.14 (January 16, 2026).
  • Counter-argument: The plugin system was added only ~3 months before the review date and has had minimal real-world validation beyond the author’s own tests. Community-contributed plugins do not yet exist, limiting the ability to assess the system’s ergonomics under diverse real-world conditions.
  • References:

Project Health Assessment

  • Evidence quality: directly verifiable (GitHub contributors, commit log, issues, releases)
  • Assessment: The project health picture is mixed. Positives: CI is solid (linting, tests, cross-platform integration tests), conventional commits are used, CHANGELOG is maintained, and semantic versioning is observed (v1.0.x). The code quality appears professional based on the structured package layout, Go Report Card badge, and golangci-lint configuration. Negatives: 19 stars and near-zero community contributions (98/100 commits from one author), 7 open issues that are all dependency bumps or minor stubs, and no releases since February 28, 2026 at review time. The project sits in a risk zone: professional enough to adopt, small enough to disappear.

Credibility Assessment

  • Author background: Kevin Elliott is a software developer. No institutional backing. The project is a personal open-source tool.
  • Publication bias: This is an open-source project README — factual and code-verifiable, not marketing material. Claims about cross-platform support and API features are checkable against the codebase.
  • Verdict: medium — The tool is functional and the README accurately describes its capabilities without exaggeration. The credibility ceiling is the small community footprint: there is no independent review, no production case study, and no external validation of the detection accuracy or API reliability. The tool works as described, but “works in tests” and “works reliably at scale” are different things.