AgentsView brings coding histories together and passes Japanese search test
AgentsView brings local conversations from coding tools into one searchable interface. A Windows 11 hands-on review found that Japanese full-text search worked, alongside chronological views of instructions, responses and tool calls. Usage charts show token consumption and estimated API costs. An experimental recall feature lets users inspect the original statements behind extracted procedures before approving or deferring them.
Artificial Intelligence··Morning
Japanese search works across local conversations
AgentsView, a tool for examining coding-agent histories, brings conversations from Claude Code, Codex, Cursor and other applications into one local interface. GIGAZINE installed it on Windows 11 and found that Japanese full-text search worked. Within a selected conversation, the timeline follows what the user requested, how the model replied and which tools it called. Users can search histories from different agents either through the search box or by pressing Ctrl+K.[1]
Usage charts separate consumption from estimated spending
The usage view groups sessions containing token information by model and day. Its costs are estimates converted to application programming interface, or API, rates. The activity view displays active and idle time and how many agents ran concurrently. A separate evaluation scores unsuccessful tool operations, repeated attempts and the load on the context. Execution times appear in an analysis view. The hands-on findings come from one GIGAZINE review.[1]
Recall links extracted procedures back to conversations
The experimental recall function turns earlier conversations into reusable procedures and points requiring caution. The user can follow that extracted knowledge to its supporting statements, then approve it or leave it pending. A Model Context Protocol, or MCP, server lets agents search prior sessions themselves. Histories from several computers can also be consolidated in PostgreSQL and analyzed with ClickHouse or DuckDB. The reviewed installation starts a local server and opens its dashboard in a browser.[1]
