Storage, Syncs, and Sturdier Analysis
This week made Humm more practical for teams with serious data requirements: customer-hosted storage, a new Postgres sync connector, richer API docs, and a cleaner path for running Agent Lab across many accounts. We also tightened the analysis loop so reports are less likely to stall on model refusals, citation issues, or awkward table output.
Customer-Hosted Data Storage
Organizations can now configure Data Sync to use their own S3 bucket instead of Humm-hosted storage. The new Data Storage settings page includes validation and copyable IAM policy instructions, so teams can keep data in their own AWS account while Humm still handles the Glue and Athena work needed to query it.
See https://heyhumm.ai/docs/guides/customer-hosted-data-sync for more details.
Postgres Data Sync
Postgres is now available as a Data Sync connector. The setup flow supports periodic replication-style sync of a Postgres database to your Humm-managed data warehouse. Supports SSH tunnel options.
More Reliable Analysis and Reports
Humm now gives models clearer authorization context before analysis, detects false refusals, retries with a targeted nudge, and falls back to another model when needed. Reflection also understands rendered data tables, artifact inventories, slash-command contracts, and citation rules more explicitly, which should reduce failed retries and make customer-specific claims easier to verify.
Public API and MCP Docs
The customer docs now separate Analyst API and Config API surfaces, use the live Config OpenAPI base URL, and include new workflow guides for managing memories, ontology, and commands through the API. MCP tool and context endpoints are also indexed in the docs, and the MCP tool surface now has REST parity for teams that prefer plain HTTP.
Improvements
- Settings > Members now supports backend pagination, search, debounced loading, and safer delete state handling.
- The Context > Ontology toolbar is cleaner: redundant sub-labels are gone, Copy moved into the main toolbar, and secondary actions now use consistent icon buttons with tooltips.
- Billing overview metrics now use clearer icons for Total Tokens and Avg. Cost / Thread.
- Analyst setup work was parallelized and warmed more effectively, reducing time-to-first-LLM setup latency in measured warm and cold runs.
- DataFrame output from sandboxed code execution is now more consistently aligned, which makes printed tables easier to read and parse.
Bug Fixes
- Fixed a Postgres sync setup crash caused by form validation assuming every credential value was a string.
- Fixed sync source health checks so Airbyte-backed sources no longer report healthy direct connections when the sync state says otherwise.
- Fixed failed report emails so the "last successful report" CTA only appears when a real previous successful report exists.