MeshAgent 0.39 brings enhancements to datasets, usage reporting, LLM proxy routing, Studio UX, and networking. The release expands dataset indexing and SQL control, completes a breaking terminology migration from databases to datasets, adds Common Crawl and Scrapy import packages, deepens billing and usage instrumentation, extends LLM proxy pipes support, modernizes mobile room navigation, and hardens network and validation paths.
MeshAgent 0.39 renames "database" terminology to "dataset" across SDKs, tooling, and CLI surfaces.
A MeshAgent dataset is the table storage inside a room. Agents and apps use it for records they need to query later: crawl results, extracted facts, task state, form submissions, retrieval tables, and other structured room data.
This is a breaking terminology migration for code that still uses older database-named APIs. New code should use the dataset-named client and toolkit surfaces.
The product change is larger than the rename. Dataset clients in Python, TypeScript, Dart, and .NET now expose stronger controls for SQL execution, index management, and table metadata. Dataset SQL calls can be cancelled, index configuration and remapping are available through the SDKs, and clients can inspect table and index metadata directly.
That matters for apps and agents that let users search, filter, import, or inspect room data. Index management gives service authors a supported way to tune the columns they search or filter. SQL cancellation lets a UI or tool stop a query that was launched with the wrong filter, is taking longer than expected, or no longer matters because the user changed context.
MeshAgent 0.39 adds two Python packages for turning web pages into room dataset rows: meshagent-commoncrawl and meshagent-scrapy.
Use Common Crawl when you want to import pages from public web archives. Use Scrapy when you want to crawl a live site now. Both paths write into a MeshAgent room dataset, report progress, and let Python callers customize how records are extracted.
The Scrapy importer can clean page content, preserve links and image metadata, keep crawl frontier state so a limited crawl can resume, and create useful dataset indexes by default. Common Crawl imports can filter URLs and customize extracted columns from WARC records.
For retrieval and research workflows, this shortens the path from web content to queryable room data. You can start from an importer that already handles page extraction, progress reporting, dataset writes, and default indexing, then customize extraction only where your product needs different columns or filtering.
MeshAgent 0.39 expands the reporting path around LLM and agent usage.
Usage records can now carry annotations, and the export path can include those annotations for cost and billing analysis. Teams can tag usage by application, service, workflow, or another internal dimension and keep that context attached when usage is exported or analyzed.
The backend reporting path also adds ClickHouse annotation and projection support, custom LLM usage tracking, and LLM tool usage instrumentation. LLM agent turn lifecycle tracing adds more detail for debugging what happened during an agent turn.
Reporting is also more visible in product and app UI. Accounts and Studio usage views can show daily usage graphs, and the Flutter Shadcn SDK now exports UsageGraph and UsageGraphPoint for apps that need interval-based usage charts.
The practical outcome is clearer ownership of usage. Admins can move from "how much did the project use?" to questions like "which app, room, user, provider, model, or workflow produced this usage?" and keep the billing context attached to the request.
MeshAgent 0.38 made proxy setup and local Codex or Claude launch flows easier. MeshAgent 0.39 extends the proxy routing path with LLM proxy pipes support, custom usage tracking, and updated pricing surfaces for gpt-5.5.
In 0.39, more proxied requests use the MeshAgent-managed routing path that attaches project, user, provider, model, and annotation metadata. That means fewer gaps in usage reports and billing exports, especially for local proxy and adapter traffic.
The CLI also exposes local proxy and logger workflows:
meshagent llm proxy
meshagent llm logger create --destination-feed-id <feed-id> --filter-expression <filter>
meshagent llm proxy runs a local MeshAgent-authenticated proxy and can show live usage while requests pass through it. meshagent llm logger routes matching LLM proxy traffic into a feed, so a team can keep an audit trail for the subset of traffic they care about.
OpenAI Responses adapter errors are clearer in this release too. When OpenAI returns insufficient_quota, MeshAgent now returns a non-retryable 402 response instead of treating the problem like a transient provider failure.
MeshAgent 0.39 adds paged response models across the Python, TypeScript, Dart, and .NET clients.
The Python client adds *_page methods for users, rooms, mailboxes, routes, feeds, OAuth clients, and scheduled tasks. TypeScript adds matching page-style methods such as listMailboxesPage, listFeedsPage, listRoutesPage, and listOAuthClientsPage. Dart adds paged models and methods for the same project and room administration surfaces, and extends paging to repositories, project secrets, services, API keys, and webhooks. .NET adds page method variants such as GetUsersInProjectPage, ListMailboxesPage, ListFeedsPage, ListUniqueUsersWithGrantsPage, and ListOAuthClientsPage.
Existing list methods now accept paging inputs such as count, offset, and filter, and the page responses include total so apps can build complete list views without guessing whether more records exist.
The CLI list commands for rooms, feeds, mailboxes, routes, and scheduled tasks also accept paging and filtering flags. That helps when a project has enough resources that dumping every row is noisy or slow.
For app builders, the change is that admin pages, dashboards, and scripts can request exactly the slice they need and show accurate pagination instead of loading an unbounded list and filtering it locally.
Studio list views also get better loading states, empty states, and table layout refinements. Account and project administration is easier to read as projects accumulate more rooms, routes, feeds, mailboxes, registries, and LLM proxy usage records.
MeshAgent 0.39 updates Studio's mobile room experience.
The mobile sidetray now prioritizes room selection, with overlay project browsing, a clearer filter focus mode, refined room selection states, empty-room handling, and fixes for scroll state during room selection. The mobile account menu moved into a flow dialog, and room transitions were hardened to avoid shared global key duplication.
Thread, voice, file, and meeting controls also get mobile polish: updated thread typography and action pills, mobile room context switcher improvements, file toolbar refinements, voice agent support in the room context switcher, and meeting control fixes for mobile surfaces.
For users who work from a phone or tablet, the app now spends less time exposing desktop assumptions. Room switching, project browsing, account actions, files, voice, and meeting controls behave more like mobile-first flows.
MeshAgent 0.39 also tightens several operational paths.
Python CLI fetches now use certifi-backed shared client sessions for more consistent TLS behavior. WebSocket and response adapter HTTP sessions now use MeshAgent's shared client-session helper instead of ad hoc aiohttp sessions. The async OAuth token exchange path in the Python CLI auth flow was fixed.
Mailbox and route domain validation helpers harden routing and mailbox configuration. That reduces the chance of accepting a route or mailbox setup that cannot work correctly once traffic reaches it.
Taken together, MeshAgent 0.39 makes datasets more consistent, usage reporting more attributable, LLM proxying easier to measure, Studio more usable on mobile and in list-heavy project views, and network behavior less surprising.
Join our Discord community to keep up with MeshAgent releases, ask questions, and share feedback with the team.
Check out the MeshAgent documentation to start building today.
MeshAgent Studio, SDK, and Server give you everything to build, test, and deploy agentic applications, from development to production.
