Vibeleaderboard
Index / app
Visit github.com
Category
Developer Tools
Rank

Previous survey · No. 794 ·

Type
APP
Builder
kenn-io
Latest release
v0.43.0
Date

About

Local-first session search, analytics, and token-usage insights for coding agents.

What it does

AgentsView gathers coding-agent transcripts into SQLite, then exposes them through a browser, command line, REST API, and assistant-facing MCP server. It can find prior conversations, inspect message context, classify projects, summarize session patterns, and estimate model costs from recorded token data.

Why it's ranked here

The tool covers the full path from transcript discovery to useful retrieval. One archive serves interactive browsing, shell scripts, API clients, and coding assistants. Read-only fallbacks, daemon-backed freshness, multiple database backends, and explicit network protections show unusual operational depth. Some advanced features require extra setup.

What's good

Search goes beyond titles and metadata: full-text, substring, regular-expression, semantic, and hybrid modes can inspect conversation content and tool output. Cost calculations account for cache creation and reads, with an offline pricing fallback. JSON output supports automation. MCP retrieval is read-only, excludes very recent sessions by default, and returns surrounding-message references.

Tradeoffs

Semantic and hybrid search require an opt-in embedding index. Git and GitHub outcome metrics are optional because the documentation calls them slow or brittle on large or missing repositories. Containers discover only explicitly mounted session directories. Remote browser and MCP access requires careful origin, bind, authentication, and transport configuration.

How to use it well

It suits developers who regularly switch among coding agents and want past solutions available from a browser, terminal, script, or current assistant. Keep the daemon running for fresh shared reads, use JSON for reporting, and add embeddings only when exact search falls short. It inspects recorded work; it does not execute coding tasks or provide writable MCP automation.

Technical notes+

The Go module is declared in go.mod; Cobra, Huma, SQLite, PostgreSQL, DuckDB, MinIO, and the Go MCP SDK appear among its direct dependencies. internal/server/server.go builds the embedded SPA and REST server around a session service. internal/mcp/server.go registers read-only retrieval tools over stdio or StreamableHTTP, retaining DNS-rebinding protection and adding constant-time bearer-token checks. internal/vector/index.go stores message and recall embeddings in separate, versioned SQLite vector schemas. internal/pricing/cmd/litellm-snapshot/main.go validates size, hash, source reference, and model contents for the compressed offline pricing snapshot. The Makefile builds the frontend into the Go binary and defines Tauri desktop bundles.

Observed

Primary language
Go
Install surface
Shell installers for macOS, Linux, and Windows; Homebrew cask; desktop releases; Docker image
Interfaces
Web UI, CLI, REST API, and read-only MCP over stdio or StreamableHTTP
Storage backends
Local SQLite archive, PostgreSQL service mode, and read-only DuckDB mirror
Desktop platforms
macOS and Windows downloads are documented; the build system also defines a Linux AppImage target
Remote session sources
Claude and Codex roots may use S3-compatible object storage

Read from README.md, go.mod, Makefile, internal/mcp/server.go, internal/vector/index.go, internal/server/server.go, internal/pricing/cmd/litellm-snapshot/main.go, docs/mcp.md, docs/data.md, docs/index.md.

Tech Stack

GoDocker

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.