↓ Skip to main content
tomrochette.com
  1. Agents/

agentsview

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 5 minutes read (~820 words)
agent-curated 173 fully-ai-generated 263 llm=glm-5.3-flash 178 observability 4 open-source 62 research-note 148 session-analytics 3 token-usage 2
Table of Contents

agentsview is a local-first, MIT-licensed Go application that discovers the session files your coding agents already write on disk, indexes them into a searchable local SQLite archive, and serves a web UI, CLI, and desktop app for browsing, analytics, and token-cost reporting across more than 60 agent formats. Facts below verified as of 2026-09-13.

agentsview’s premise is that your agents already write the telemetry; the missing piece was a tool that reads all of it in one place, locally, instead of each harness showing you only its own slice.

What it is
#

A Go daemon that watches known per-agent session directories (Claude Code, Codex, Gemini CLI, Copilot, Cursor, Zed, Windsurf, OpenCode, Qwen Code, Goose, Kiro, and more, more than 60 sources listed), parses their JSONL and database logs, and syncs them into a local SQLite database with FTS5 full-text search. Surfaces: a web UI on loopback, a CLI (agentsview usage daily, session search, stats), a Tauri desktop app, Docker, plus optional PostgreSQL push for shared team dashboards, S3-backed session roots, and a DuckDB mirror. Token-cost reporting uses a model-pricing catalog; an earlier version of the docs benchmarked reports at 84 to 223 times faster than re-parsing with ccusage (called an upper bound by the docs themselves), but the current docs have dropped that benchmark and now simply note ccusage covers the same core job. Pure local file parsing: no cloud service, no accounts, no LLM calls required; an anonymous activity ping fires by default and can be disabled. Made by Kenn Software LLC; install via curl script, Homebrew cask, desktop builds, or Docker.

Status
#

Young and active: about 5.9k stars, 663 forks, 114 open issues and PRs as of 2026-09-13, created 2026-02-19. Latest release v0.42.0 on 2026-09-01 with roughly weekly releases since July; pre-1.0 with fast feature churn.

Strengths
#

  • Breadth is the product: one tool indexes every major agent’s sessions, where most alternatives are Claude-only.
  • A pre-indexed store makes multi-month, multi-agent cost and history questions answerable in seconds instead of ad-hoc parsing.
  • Serious team story: PostgreSQL push, machine-labeled filesystem sync, S3 roots, and versioned export schemas.
  • Careful local-first hygiene: loopback binding, DNS-rebinding guards, auth for non-loopback exposure, and a documented trust model for untrusted session files.

Cautions
#

  • Pre-1.0 with breaking schema bumps (usage output is at schema version 6 in the docs’ JSON contract), so scripts consuming its output churn.
  • Token coverage is opportunistic: cost rows appear only when a transcript contains both token counts and a priceable model, and the docs admit known undercounts such as Claude WebSearch side-calls.
  • The 84-223x benchmark numbers the docs used to publish were vendor-supplied on vendor hardware, and their removal leaves no performance evidence in either direction.
  • The daemon, serve, DuckDB, and PostgreSQL surface is real operational weight if all you wanted was a cost report.

Pricing
#

Free and open source under MIT, no accounts, everything local. No paid tiers are published.

Compared to
#

  • Built-in harness commands (Claude Code’s usage and cost views): zero-install and authoritative for the current session, but single-agent, single-machine, and history-free; choose agentsview for retrospective multi-agent analytics.
  • simple10/agents-observe: a Claude Code plugin capturing live hook events for in-flight dashboards; choose it for what my agent is doing right now, agentsview for a passive index of everything already written.
  • xintaofei/codeg: a full multi-agent workspace where aggregation is a byproduct of running agents inside it; choose agentsview when you keep your existing setups and want a read-only analytics layer.

Bottom line
#

Recommended for engineers running three or more different coding agents who want one private place to answer what did my agents do and cost this month. Not for single-agent users well served by built-in cost views, or teams wanting live in-flight observability rather than retrospective indexing.

Changes
#

  • 2026-08-30 - Created in the new Session analytics category as a research note on the local session indexer.
  • 2026-09-05 - Updated the supported-formats claim to 60-plus agent formats, up from roughly 55, including the matrix cell.
  • 2026-09-08 - Revised prose, caution, and references after the docs dropped the 84-223x ccusage benchmark entirely.
  • 2026-09-13 - The docs’ usage JSON contract now documents schema version 6 (provider-specific billing identity), so the schema caution moved from version 5 to 6.

See also
#

References
#