Five-minute guide
The session intelligence loop
Your agents already write the raw data. AgentsView turns it into a searchable archive, and the archive into answers your agents can use on their next run.
-
Capture every session
The daemon discovers the session directories behind more than 60 agent formats, including Claude Code, Codex, Cursor, Copilot, and Gemini, and syncs them into one local SQLite archive with full-text indexes. It keeps watching, so the archive stays current while you work.
Quick start
The dashboard after first sync. -
Browse the full conversation
Read the prompts, responses, reasoning, and tool calls recorded by each agent. Filter by project, agent, date, and message count, or press Ctrl/Cmd+G to open a session by ID. The Resume menu lets supported agents, including Pi, pick up where they left off.
Use the opt-in project workspace to correct project names with folder suggestions, transcript previews, and bulk corrections.
Usage guide
A transcript with tool calls and thinking blocks. -
Monitor the fleet
The Activity dashboard shows when agents ran, how much work overlapped, and what it cost: peak concurrency with the exact moment it happened, active versus idle time, and agent-minutes over any window. Click a timeline bucket to see exactly which sessions were running in that slot.
Activity reference
A week of agent activity and concurrency. -
Meter tokens and cost
Usage reports show recorded tokens and estimated costs, including prompt-cache writes and reads. Reports reuse saved archive data; the first request after an upgrade or sync may need time to prepare its cache. The CLI answers in the terminal, the statusline shows today's spend inside your editor, and one-shot capture reports recorded usage for a single Claude or Codex CI run.
Token usage and costs$ agentsview usage daily --breakdown $ agentsview usage statusline $9.61 today
Daily token and cost breakdown. -
Search by words or by meaning
Press Ctrl/Cmd+K to search, then choose a project and date range in the palette. Select All Projects to widen the search without changing the sidebar. Full-text search finds exact words; opt-in semantic and hybrid search find related meaning and cite the matching conversation.
Search controls · Semantic search
Choose a project and date range for your search. -
Assess session health
Health scores point to tool failures, context pressure, and repeated loops in the transcript. Open Analysis to see Session Vitals: measured tool execution and time that cannot be assigned to a phase. Click an activity row to inspect its transcript entry.
Session intelligence
Measured tool execution and unattributed turn time. -
Keep what the sessions learned
Recall (experimental) collects reusable lessons from your archive. Browse each entry and follow its evidence links to the source messages. Generated Insights add model-written reports over an explicit session scope.
Recall reference
Generated insights over a session scope. -
Give your agents the archive
The loop closes when agents read their own history. The MCP server exposes session history as assistant tools, the REST API and CLI serve scripts and hooks, and SSE streams live messages. An agent can check what a previous run tried before repeating it.
MCP server · Session API
-
Extend beyond one machine
Push each machine's archive to PostgreSQL for a merged team view, use ClickHouse for a remote dashboard, mirror into DuckDB for analytical queries, read source files through the filesystem or S3, or keep original files in hosted storage. SQLite on your disk remains the local archive of record.
PostgreSQL sync · ClickHouse sync · DuckDB mirror
Two machines merged in one session list.
Next
Run it on your own history.
Install AgentsView and import the sessions already on your machine. The docs cover configuration, exact command behavior, and the full feature reference.