AgentsView0.46.1

AgentsView brings past AI coding sessions into one searchable archive, allowing developers to browse conversations, inspect what agents did, and revisit earlier work. It imports supported records from more than 60 agent formats, so users can compare activity, costs, and results across projects in one place. A background server watches the directories that coding agents already use, detects their default locations automatically, and can also monitor locations you configure. New messages can stream into the interface while a session is still running.

The Activity dashboard reports peak concurrency, the precise time it occurred, active and idle periods, agent-minutes across parallel sessions, and spending. Users can narrow those reports by day, week, month, or a custom period, then filter by project, agent, or machine. Cost reports use the indexed archive, follow LiteLLM and OpenRouter rates, support offline pricing fallback, and account for prompt-cache creation and reads.

Full-text search finds terms in transcripts, while optional semantic and hybrid search find conversations by meaning; the command palette searches by project and date, and Ctrl/Cmd+G opens a session by ID. Health scores and quality indicators link back to supporting transcript evidence, while experimental Recall extracts decisions, warnings, and project facts with source-message links. Generated Insights create model-authored reports for a defined session scope, and the archive can be accessed by agents through the CLI, REST API, or MCP server.

The archive is local by default and remains on the machine unless a sharing feature is enabled; storage uses SQLite, with PostgreSQL, DuckDB, and ClickHouse support. This macOS Developer Tools release targets Apple Silicon (arm64), with Linux and Windows availability plus desktop, pip, and Docker forms, and is freeware under the MIT license. It suits developers reviewing AI-assisted coding history, measuring agent usage and costs, or giving agents access to their own prior attempts.