v0.3.3

What's Changed

Features

• Sidebar conversation filter — filter conversations directly from the sidebar (#59)
• Tab autocomplete for agent name — tab cycles through agent names during onboarding (#58)

Fixes

• Backend-driven display fields — engine connection forms now use server-defined fields (#54)
• CLI log tailing — tail logs no longer requires an external tail command (#60)

Chores

• Added missing license file (#55)

Full Changelog: v0.3.1...v0.3.3

v0.3.2

What's Changed

Features

• Sidebar conversation filter — filter conversations directly from the sidebar (#59)
• Tab autocomplete for agent name — tab cycles through agent names during onboarding (#58)

Fixes

• CLI log tailing — tail logs no longer requires an external tail command (#60)
• Backend-driven display fields — engine connection forms now use server-defined fields (#54)

Chores

• Added missing license file (#55)

Full Changelog: v0.3.1...v0.3.2

v0.3.1

What's new

Custom authentication headers for OpenAI-compatible endpoints

The OpenAI-compatible provider now supports arbitrary HTTP request headers, making it easy to connect to LLM gateways that require proprietary tokens or custom auth schemes — not just a plain API key.

• Add any number of key-value headers from the Model Settings panel or the onboarding wizard
• Header values are encrypted at rest and redacted in the UI — keys are shown on reload so you know what's saved
• Existing OPENAI_COMPAT_BASE_URL / OPENAI_COMPAT_API_KEY env var setups continue to work with no changes

Other improvements

• Verify timeout — LLM connection tests now time out after 30 seconds with a clear error message instead of hanging indefinitely
• Better error messages — error details from the agent are now unwrapped correctly and shown in the chat UI

Upgrade

pip install --upgrade datahub-analytics-agent

Or with uv:

uv tool upgrade datahub-analytics-agent

v0.3.0

What's new

Hive / Kyuubi / Spark Thrift Server connector

Analytics Agent now supports Hive-compatible databases via a new native connector. Connect to HiveServer2, Apache Kyuubi, or Spark Thrift Server from the Add data source UI — the connector installs on demand, no manual setup needed.

Supports NONE, NOSASL, LDAP, and KERBEROS auth modes. Kerberos service name is configurable via the UI.

Bug fixes

• list_tables validation error — fixed a crash where the agent passed `schema=null` to the `list_tables` tool, causing a Pydantic validation error. Affected Snowflake, BigQuery, and Hive connectors (#48)
• crypto.randomUUID on non-HTTPS deployments — fixed a `TypeError` that occurred when the app was served over plain HTTP on an internal host. Replaced with the `uuid` package which works in all contexts (#47)

Upgrade

pip install --upgrade datahub-analytics-agent

Or with uv:

uv tool upgrade datahub-analytics-agent

v0.2.3

What's New

Update notifications

• Amber pulsing dot on the Settings gear icon when a newer version is available
• Settings opens directly to the About tab when an update is detected
• About tab shows installed vs. latest version with a one-liner upgrade command (analytics-agent upgrade)
• Full GitHub changelog viewer — release cards with expandable notes, installed release highlighted

analytics-agent upgrade CLI

• New upgrade [--to VERSION] command to update via pip
• Clear guidance for editable (repo) and uvx installs instead of a broken pip run

OpenAI-compatible proxy support

• Connect any OpenAI-compatible endpoint (LiteLLM, vLLM, Ollama) as the LLM backend

Installation / Upgrade

pip install --upgrade datahub-analytics-agent
# or
analytics-agent upgrade

v0.2.2

Bug fixes

pip install / uvx now works end-to-end — two issues that broke the no-clone install path (introduced in v0.2.1) are resolved:

• Blank UI after install — analytics-agent quickstart would start successfully but the browser returned 404 Not Found. The published wheel was missing the frontend assets because uv build built via an sdist intermediary, which excluded the gitignored frontend/dist/ directory. Fixed by building the wheel directly.

• uvx datahub-analytics-agent quickstart failed — uvx expects the executable name to match the package name. The datahub-analytics-agent entry point is now registered alongside the existing analytics-agent alias, so both work.

If you installed v0.2.1 via pip or uvx, upgrade to get a working UI:

pip install --upgrade datahub-analytics-agent
# or
uvx datahub-analytics-agent@latest quickstart

New Contributors

• @lakshay-nasa made their first contribution in #29

Full Changelog: v0.2.1...v0.2.2

v0.2.1

Zero-install quickstart

You can now go from nothing to a running agent in one command — no repo clone required:

pip install datahub-analytics-agent && analytics-agent quickstart
# or: uvx datahub-analytics-agent quickstart

The interactive wizard prompts for your LLM provider and API key, an optional data source, then bootstraps the database, starts the server, and opens the browser. Config lives in ~/.datahub/analytics-agent/ alongside the DataHub CLI.

Additional CLI commands: start, stop, status, logs, config.

Bug fixes

• Fixed a silent bootstrap failure on fresh installs: analytics-agent bootstrap was a no-op when invoked via python -m (missing __main__ entry point), leaving the database empty and the server unable to start
• Fixed Alembic migration discovery for pip-installed users: migrations now locate scripts via the installed package path rather than requiring alembic.ini in the working directory
• Fixed Docker image build: hatch_build.py is now included in the image so uv sync succeeds without the source repo present

v0.2.0

BigQuery query engine

BigQuery is now a first-class query engine alongside Snowflake and DuckDB. Configure it through the settings UI; the agent validates credentials and project access at connection time and surfaces actionable errors (e.g. when a project ID is set but credentials are missing).

Helm chart and bootstrap CLI

A production-ready Helm chart is available under helm/ for Kubernetes deployments. The new analytics-agent bootstrap CLI command handles database migrations and seeds config.yaml on first run and after upgrades — the recommended entry point for all deployment methods.

Prompt caching and token usage

Prompt caching is enabled by default for Anthropic and AWS Bedrock providers, reducing cost on repeated context. Per-call token counts and session totals are now displayed in the UI.

MCP subprocess architecture for connectors

Connector dependencies (e.g., BigQuery and Snowflake client libraries) now run in an isolated subprocess via the Model Context Protocol. This eliminates version conflicts between connector packages and keeps the main process dependency footprint small.

Charts rendered outside the work block

Query result charts are now displayed at the top level of the response rather than inside the collapsible reasoning trace, so results are immediately visible without expanding the work block.

Bug fixes

• Fixed execute_sql appending LIMIT to non-SELECT statements (DDL, CTEs, etc.)
• Fixed blocking DataHub API calls in async request handlers, which caused latency spikes under concurrent load
• Fixed quickstart token provisioning and AWS credentials plumbing in the bootstrap flow

---

Full Changelog: v0.1.0...v0.2.0

v0.1.0

What's Changed

• chore: rename package to datahub-analytics-agent + add PyPI metadata by @shirshanka in #2
• docs: add node and just to manual setup prerequisites by @tazmeen24 in #3
• chore: add PyPI publish workflow via OIDC trusted publishing by @shirshanka in #4
• feat: add AWS Bedrock provider + bump langchain to 1.x by @nwadams in #6
• fix: abort in-flight SSE stream when switching conversations by @shirshanka in #8
• feat: per-call token usage tracking with session totals by @nwadams in #7
• chore: change default ports to 8100 (prod) and 8101 (dev) by @shirshanka in #9

New Contributors

• @shirshanka made their first contribution in #2
• @tazmeen24 made their first contribution in #3
• @nwadams made their first contribution in #6

Full Changelog: https://github.com/datahub-project/analytics-agent/commits/v0.1.0