Agentsview is a local-first analytics and monitoring platform for AI coding agents, written in Go and available as a desktop application, CLI tool, or Docker container. The project enables developers to browse, search, and track costs across more than 20 supported coding agents including Claude Code, Codex, Devin, and others, all without requiring cloud accounts or external services. The core philosophy centers on keeping all data local, with a single binary deployment model and a web UI served at localhost by default.
The repository has grown modestly in recent activity, gaining 2 stargazers between July 5 and the present check, reaching 3741 total stars. Issue and pull request response latency shows a median of 9.5 hours with a mean of 243.9 hours across 228 tracked items, indicating variable response times. The most active contributor tracked by GitGenius is wesm with 315 events, followed by mariusvniekerk with 29 events and tpn with 15 events. Bug reports and enhancement requests dominate the issue tracker with 11 and 10 labeled items respectively. The project shares contributors with notable repositories including pandas-dev/pandas, cockroachdb/cockroach, and gastownhall/beads, suggesting involvement from developers working on large-scale data and infrastructure projects.
Agentsview's primary feature set revolves around session discovery and local storage. On first run, the application automatically discovers sessions from every supported agent on a user's machine and syncs them into a local SQLite database. The web dashboard provides full-text search across message content, token usage and cost tracking with per-session and per-model breakdowns, activity heatmaps, and tool usage analytics. The interface emphasizes keyboard-first navigation with shortcuts like j/k for movement and Cmd+K for search. Sessions can be exported as HTML or published to GitHub Gist.
Token usage tracking is a core capability, with the agentsview usage command providing fast local cost analysis across all agents. The tool calculates costs using LiteLLM rates with offline fallback and includes prompt-caching-aware calculations that distinguish between cache creation and read tokens. Users can filter by date range, specific agents, and request per-model breakdowns or JSON output for scripting. The per-session usage command reports total output tokens, peak context tokens, and USD cost estimates when pricing data is available.
The platform supports multiple deployment modes. Desktop applications are available for macOS and Windows via GitHub Releases or Homebrew. A Docker image is provided with examples for PostgreSQL-backed and DuckDB-mirrored configurations. For remote access through SSH port-forwarding or reverse proxies, the server validates Host headers to prevent DNS-rebinding attacks and supports public-url configuration with optional authentication requirements.
Session data ingestion is flexible and security-conscious. For local agents like Devin CLI, agentsview reads session data from configured directories while intentionally ignoring copied config files and OAuth paths. For Claude and Codex, sources can be configured as S3-compatible object storage roots, allowing central instances to read sessions from multiple machines. S3 change detection uses size, modified time, and available fingerprints like ETag to minimize bandwidth during sync operations.
The analytics dashboard provides window-scoped statistics including session totals, archetype classifications, distributions for duration and context size, cache economics, and temporal hourly breakdowns. Git-derived metrics like commits and lines of code are opt-in due to potential performance concerns on large repositories. The Recent Edits feed aggregates files changed by agents across all sessions, grouped by project and path with links to the originating messages.
GitGenius classifies agentsview across multiple domains including agent monitoring, performance metrics, real-time dashboards, system analytics, resource usage tracking, and data visualization, reflecting its comprehensive approach to coding agent observability.
