AgentsView Session Archive
Search coding-agent sessions locally and inspect token use, estimated costs, and activity.
- Source repo
- kenn-io/agentsview
- Stars
- ★ 6.1k
- Last updated
- today
- License
- MIT
- Primary language
- Go
- FA score
- 68/100 · Some gaps
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 using Claude Code, Codex, and other tools wants one local interface to search and revisit sessions.
- Not a fit if
- Developers who need session browsing inside their IDE
- Teams that need Cursor attribution stats synced to PostgreSQL
- Source review
- 68/100 · Some gaps
What does this agent do, and when should you use it?
AgentsView is a local-first session browser and analytics tool for coding agents, with a Go CLI, a Go-served web UI, and desktop apps for macOS and Windows. On first run it discovers sessions from supported agent directories, syncs them into a local SQLite archive, and serves the UI at 127.0.0.1:8080 by default. The CLI lists and searches sessions, reports token use and estimated costs, and emits windowed statistics; the UI provides full-text search, session viewing, cost dashboards, and activity analytics. Optional PostgreSQL, ClickHouse, and DuckDB targets can receive pushed or mirrored archive data and serve it read-only. Session content stays local by default; telemetry can be disabled, while semantic search, GitHub Gist export, and remote sync involve their configured external services.
AgentsView reads local session files from Claude Code, Codex, and other supported tools, parses them, and stores sessions and messages in SQLite. agentsview serve starts the server, web UI, sync, and file watchers; agentsview daemon start runs it in the background. Commands such as agentsview session list, agentsview usage daily, and agentsview stats read the archive and return sessions, cost summaries, and analytics. agentsview session search supports full-text search, with semantic or hybrid search available after configuring an embeddings endpoint. The web UI displays transcripts, token use, costs, activity heatmaps, and tool usage; the REST API exposes per-session usage at /api/v1/sessions/{id}/usage. Optional pg push, clickhouse push, and duckdb push copy the SQLite archive to those destinations for read-only serving or analysis.
- A developer using Claude Code, Codex, and other tools wants one local interface to search and revisit sessions.
- An individual developer wants daily or per-model token and estimated-cost reports across coding agents.
- A team wants to push local archives to PostgreSQL for a shared, read-only dashboard.
- An analyst wants a DuckDB mirror or a ClickHouse copy for read-only querying.
- A user building automation wants CLI JSON output, stats, or the per-session REST API for scripts.
How do you install or deploy this agent?
Install the CLI on macOS or Linux with:
curl -fsSL https://agentsview.io/install.sh | bashOn Windows, use PowerShell:
powershell -ExecutionPolicy ByPass -c "irm https://agentsview.io/install.ps1 | iex"The desktop app is available for macOS and Windows; on macOS, Homebrew is also supported:
brew install --cask agentsviewThe Docker image is another documented option. Mount agent session directories and set their matching environment variables so the container can discover those sessions:
docker run --rm -p 127.0.0.1:8080:8080 \
-v agentsview-data:/data \
-v "$HOME/.claude/projects:/agents/claude:ro" \
-e CLAUDE_PROJECTS_DIR=/agents/claude \
ghcr.io/kenn-io/agentsview:latestHow do you use this agent?
On first run, AgentsView discovers supported session sources, syncs them into SQLite, and starts the local web UI:
agentsview serveOpen http://127.0.0.1:8080 in a browser. Common CLI commands include:
agentsview daemon start
agentsview session list
agentsview usage daily
agentsview stats --format jsonSearch with agentsview session search; semantic search requires an OpenAI-compatible embeddings endpoint. Per-session usage is available from the CLI or REST API:
agentsview session usage <id>
curl http://127.0.0.1:8080/api/v1/sessions/<id>/usageDaily usage reports read the saved archive. Run agentsview sync first to include new changes from source directories.
What are this agent's strengths and limitations?
- Consolidates local sessions from more than 20 coding agents in SQLite, with FTS5 full-text search and project/date filters.
- Provides cross-agent usage, cost, and analytics through CLI reports, JSON output, and a per-session REST API.
- Keeps SQLite as the primary archive while offering PostgreSQL, ClickHouse, and DuckDB as optional push or read-only mirror backends.
- Offers CLI, web UI, macOS and Windows desktop apps, and Docker deployment.
- Each agent needs an accessible local session directory; Docker deployments must mount directories and set matching environment variables.
- Semantic search requires an OpenAI-compatible embeddings endpoint; cost estimates depend on available pricing, and some proprietary models may be unpriced.
- PostgreSQL, ClickHouse, and DuckDB features require extra backend configuration; remote access also requires appropriate authentication, TLS, or a trusted proxy.
- The README does not specify desktop OS version requirements; source development requires Go 1.27+ (CGO) and Node.js 24.11+.
How does this agent compare with similar options?
The README presents SQLite as the primary archive and PostgreSQL, ClickHouse, and DuckDB as optional shared or mirror backends. SQLite supports writes and FTS5 indexed search; remote backends serve read-only. DuckDB search uses substring or regex fallback, while SQLite FTS5 is the indexed search path.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AgentsView Session Archive This agent | 68 · Some gaps | CLIFree | ★ 6.1k | today | Go | Codex · Claude Code |
| LeanCTX Context Intelligence | 75 · Good | CLIFree + model costs | ★ 3.8k | 12d ago | Rust | Codex · Claude Code · OpenAI API · Claude API |
| Agent Deck | 76 · Good | CLIFree + model costs | ★ 1k | 2d ago | Go | Codex · Claude Code |
| ccxray | 73 · Some gaps | CLIFree + model costs | ★ 296 | 9d ago | JavaScript | Claude Code · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The repository describes a loopback default, Host/CORS/CSP defenses, authentication boundaries, remote sync, and outbound connections. It says non-loopback PostgreSQL requires TLS by default and that LiteLLM pricing and update requests send no session content. It also discloses default secret redaction, plaintext storage, and deletion limits. Deductions: the archive remains plaintext with permissions inherited from umask, detection is not complete data-loss prevention, and some exports, syncs, and deletion effects are not reliably reversible; confirmation is not established for risky operations generally. A dependency manifest is present but no vulnerability audit evidence is supplied. Publisher identity is unverified, and the private vulnerability reporting channel remains TODO.
The README explains daemon, serve, sync, and backend behavior in useful detail, while the security document states deployment assumptions. Deductions: the supplied evidence does not establish general dependency availability or offline behavior beyond a LiteLLM pricing fallback, and failure messaging and recovery paths are thin. CI configuration shows focused test commands but does not establish that they ran or passed.
The materials target single-user developer workstations and describe many coding agents, platform paths, Docker, SSH forwarding, remote deployment, and one-shot CI capture. Deductions: shared-host use, multi-user isolation, and strong deletion guarantees are explicitly out of scope; some capabilities have prerequisites, such as explicit Aider configuration and deprecated Amp support, with further configuration details delegated to external documentation.
The README is organized around installation, quick start, remote access, usage, browsing, and supported agents, and includes command examples; the MIT license is complete. Deductions: installation spans platforms but includes remote script piping without checksum or signature verification instructions in the supplied files. A version and changelog entry point are given, but the README says documentation follows main, so they are not fully pinned to this revision. Maintenance responsibility and a private security contact are unclear.
Search, cost reporting, analytics, cross-agent session browsing, and JSON output provide concrete utility for coding-agent archives, and semantic search is identified as opt-in. Deductions: indexing and retaining session content creates substantial sensitivity, while operating a daemon, data backends, and optional sync adds setup and upkeep costs. Some outcome metrics require optional Git/GitHub queries that the README says may be slow or brittle.
The README gives checkable commands, paths, API fields, and security boundaries; SECURITY.md separates storage, outbound traffic, and defense assumptions; the CI workflow lists specific test commands. Deductions: many feature and security claims remain documentation assertions because implementation files are not supplied; CI configuration is not evidence of passing tests; and the security document calls itself foundational, with some decisions left open.
- The archive can contain prompts, tool arguments, command output, and secrets, and is stored in plaintext; restrict access to its data directory and treat exports, remote sync, and external databases as additional copies.
- Configure authentication and trusted origins carefully when enabling non-local access; non-loopback plaintext Quack and PostgreSQL allow_insecure weaken transport protection.
- Secret detection is best-effort, and permanent deletion does not guarantee removal from WAL files, SQLite pages, backups, or external copies. The private security reporting channel is not yet configured.
FAQ
Do I need a paid model or API key to use AgentsView?
Does session content get uploaded to the cloud?
AGENTSVIEW_TELEMETRY_ENABLED=0 or TELEMETRY_ENABLED=0; shared backends, semantic-search endpoints, and Gist export are optional external paths.Why is an agent or session missing from the UI?
AIDER_DIR or agents.aider.dirs; Docker also needs the source directory mounted.Will cost figures exactly match billing?
Can I open the local server through a remote browser?
--public-url or trusted origins for forwarding or proxy access. If exposing the UI beyond loopback, enable --require-auth; these settings do not create an external listener or proxy.