v2.7.1

Changed

• Remote callers no longer receive your configuration paths or the file paths of documents they change. A REST response to a caller that is not on your machine (the same localClient rule as document paths) and every Streamable HTTP MCP result now omit the config file and index database locations (configPath, dbPath) and each collection's root folder (path) from /api/status, /api/collections, and gno_status. REST /api/status also omits suggested-folder, model cache, and model file paths, and /api/connectors omits connector install paths. Document create, editable copy, rename, move, duplicate, trash, save, and create-folder responses also omit the file's host path (and the save's file:// uri) for remote callers; they still carry the gno:// URI and relPath. The dashboard's disk check now reports free space for the model cache without naming its folder. Remote callers name collections by name. The Collections page, dashboard, and connector list work as before on your machine; a remote Web UI shows them without the path lines. path, configPath, and dbPath are now optional in the status and collection-list schemas. See Host Paths and Remote Callers.

Fixed

• Windows: writes with a request ID (gno capture --request-id, gno remember --request-id, and the MCP, REST, and SDK equivalents) no longer start PowerShell on every call to check the private ledger directory's permissions. The check runs once and is recorded, and later calls skip it while the directory and its permissions are unchanged. A replaced directory or a permissions change triggers the full check again, and a directory another account can access is still refused before anything is written.
• Windows: a first request-ID write interrupted before the private ledger directory was secured (for example by a slow PowerShell start) no longer leaves every later request-ID write refused. An empty ledger directory is now secured like a new one, and the permission check allows 30 seconds for PowerShell to start.
• gno serve and gno daemon now drain an embedding backlog that already exists when they start (for example after gno index --no-embed or an interrupted run) instead of waiting for the next file change; the first pass runs 30 seconds after startup.
• A background embedding batch that exceeds models.inferenceTimeout no longer aborts the whole pass. Only that batch fails and stays pending; later batches still embed, so one slow chunk at the head of the backlog cannot keep every other chunk pending.
• Agent sessions: a running gno serve on a session archive picks up sources added or removed with gno sessions source add/remove without a restart, both on the Sessions page and in /api/sessions/status. Remove on the page succeeds for a source the CLI already removed, and a config file the server cannot read is reported instead of serving stale sources.
• Agent sessions: pressing Enter on "Discover local sources" moves keyboard focus to the discovery results instead of dropping it.
• URIs from a named index now carry ?index= in gno ls --json and in gno ask --json meta.answerContext, like the other JSON outputs, so they read back from the right index with gno get.
• The session archive guide explains how the pending unit count relates to the other unit counts in gno sessions status.
• Session import and other record-collection syncs (JSONL, mail, calendar exports) now take time in proportion to the number of records. Before, each new record scanned the whole index, so large histories slowed down quadratically. A synthetic history of 90 threads with 200 turns each now imports in about 47 seconds instead of 27 minutes, and 300 threads take about 3 minutes. Index contents, import receipts and checkpoints are unchanged.