🧠 Run Evaluations via SDK

Execute Tests for Agents directly through the SDK — no UI required.
Validate the latest configuration, automate regression checks, and plug evaluations into CI/CD.

⚙️ Full Dataset CRUD via SDK

Manage datasets programmatically with full create, read, update, delete support.
Mirror everything you can do in the UI—now scriptable for automation and integrations.

🔍 Grounding Prompt Now a Core Prompt

The Grounding (llm_fuzzy_match) prompt is now in the Core Prompt Library.
View, edit, and trace how value matching happens (e.g., “NY” → “New York”), and pair changes with evaluation sets for quality guardrails.

🧮 Parameter Prompt Visibility & Control

Bring transparency to skill parameters (e.g., filters, metric selections).
Use Jinja templating to override default text and precisely control how parameter prompts are described.

📅 Auto-Generate Max Date Fields

During dataset creation, Max now auto-creates max_date fields to power SQL Gen that depends on date dimensions.
Generate on demand from the dataset context menu; SQL is tailored to your connection’s dialect.

🧪 Choose Evaluation Sets for Prompt Tests

In the Prompt Library, pick which evaluation sets (collections) to run.
Target specific requirements and save time and cost by avoiding all-collections runs.

👥 Question Tests by User Context

Run question tests as a specific user to reflect real permissions, data visibility, and personalization.
What you test now matches what users actually see.

📦 Selectable Dependencies When Exporting Agents

Choose which dependencies to include when exporting an agent (connections, datasets, skills).
Lighter exports, cleaner migrations, fewer surprises.

🧭 Control Researcher’s Final Message Prompt

The final message prompt used by the Researcher is now configurable in the Prompt Library.
Developers can edit or override this prompt to fine-tune how research findings are summarized and communicated to users.

📊 Chart Conversions

In Chat and in the visualization builder, toggle between chart types without rebuilding.
Max auto-transposes data as needed—explore the same result through multiple lenses in a click.

📋 Table Row Collapse Control

Set whether rows in table visualizations load collapsed or expanded by default.
A per-table setting for cleaner first looks or deeper immediate detail.

📄 Custom PDF Downloads

Define purpose-built layouts for PDF downloads to avoid awkward page breaks and layout drift.
If no custom layout is provided, we fall back to exporting the chat view as before.

👤 User Info Panel: Full-Page View

The user info experience moves from a side panel to a dedicated full page.
A small tweak that brings consistency and simplicity to the UI.

⏰ Schedule Panel Relocation

Agent schedules now live in Agent Settings (moved from the Chat side panel).
All agent controls, one place—more consistent configuration.

⚠️ Deprecation Notice: Skill/Global Functions Removed from Code Skills

Skill Functions and Global Functions are no longer supported in Code Skills.
Code Skills should define shared logic within the skill’s repository (standard modules and imports).

This change does not affect existing component-based skills, which continue to support Skill and Global Functions.

Additional Note for Self-Hosted Customers

This release does not introduce additional upgrade requirements beyond standard processes.
As always, we recommend running backups before upgrades.

To download Max from our container registry use the tag: v25.11.0

Tested Models

AnswerRocket SDK version: 0.2.95

OpenAI:

Chat: gpt-4.1-2025-04-14
Narrative: gpt-4.1-2025-04-14
Evaluations: gpt-4.1-2025-04-14
SQL: o4-mini-2025-04-16
Researcher: gpt-5-mini-2025-08-07
Embedding: text-embedding-3-small

Gemini:

Chat: gemini-2.5-pro
Narrative: gemini-2.5-pro
Evaluations: gemini-2.5-pro
Embedding: models/embedding-001

Anthropic:

Chat: claude-sonnet-4-5-20250929
Narrative: claude-sonnet-4-5-20250929
Evaluations: claude-sonnet-4-5-20250929
Embedding: Not Applicable