unique-deep-research 3.5.0

Deep Research Tool for complex research tasks

Deep Research Tool

Deep Research Tool for complex research tasks that require in-depth investigation across multiple sources.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[3.5.0] - 2026-02-27

• Add support for exporting the report to a Word Document

[3.4.1] - 2026-02-23

• Add flag-based cancellation support via CancellationWatcher

[3.4.0] - 2026-02-20

• Add option to show full page content of search results in the return of the web search tool

[3.3.2] - 2026-02-11

• Correctly log debug info in case of error

[3.3.1] - 2026-02-09

• Migrate to model-specific token counting from unique_toolkit 1.46.1

[3.3.0] - 2026-02-06

• Expose configuration for custom engine config (e.g max_parallel researchers)
• Makes prompts aware of the available tools
• Add option to disable the web_fetch tool

[3.2.3] - 2026-02-05

• Improve error handling and logging

[3.2.2] - 2026-02-05

• Fix bug where deep research tool was not using the correct headers causing an authentication error

[3.2.1] - 2026-02-05

• Use deep research logger instead of tool logger

[3.2.0] - 2026-02-03

• Use a backwards compatible config style

[3.1.0] - 2026-01-30

• Support other search engines than Google

[3.0.28] - 2026-01-26

• Specify langchain-openai version

[3.0.27] - 2026-01-23

• Fix bug where deep research model usage analytics were not fully exported

[3.0.26] - 2026-01-16

• Add local CI testing commands via poethepoet (poe lint, poe test, poe ci-typecheck, etc.)

[3.0.25] - 2026-01-16

• Add unified type checking CI with basedpyright

[3.0.24] - 2026-01-15

• Add pytest-cov dev dependency for coverage testing

[3.0.23] - 2026-01-13

• Add missing pytest-asyncio dev dependency
• Fix tests to use .invoke() method for StructuredTool objects

[3.0.22] - 2026-01-13

• Fixing bug with message logs

[3.0.21] - 2025-12-29

• Bump unique_sdk version to 0.10.58

[3.0.20] - 2025-12-17

• Bump unique_toolkit version to 1.38.3

[3.0.19] - 2025-12-17

• Update failsafe execution import path

[3.0.18] - 2025-12-04

• Change tool name in logs to DeepResearch instead of Deep Research

[3.0.17] - 2025-12-03

• Include tool usage log into debug info

[3.0.16] - 2025-12-01

• Upgrade langchain libraries to ensure compatibility with websearch

[3.0.15] - 2025-11-12

• Move pytest to dev dependencies

[3.0.14] - 2025-11-10

• Include a check to ensure that the website provided by OpenAI in the event is not None
• Wrap OpenAI event processing in try catch block to better handle unexpected types

[3.0.13] - 2025-11-07

• Include ability to toogle enabled tools for deep research
• Standardize web search message log

[3.0.12] - 2025-11-06

• Include pytest test suite

[3.0.11] - 2025-11-05

• Include chat_id and user_id in headers sent to openai proxy endpoint

[3.0.10] - 2025-10-31

• Make agent aware of limitation in data access
• Apply metadata filter to agent

[3.0.9] - 2025-10-31

• Clear original response message when starting new run
• Forced tool calls fix for setting research completed at max iterations
• Reduce web search results returned to 10 to reduce api load

[3.0.8] - 2025-10-29

• Include DeepResearch Bench results

[3.0.7] - 2025-10-28

• Removing unused tool specific get_tool_call_result_for_loop_history function

[3.0.6] - 2025-10-20

• Include find on website events in message log

[3.0.5] - 2025-10-17

• Add all reviewed sources to message log

[3.0.4] - 2025-10-14

• Fix ordering issue of messages in unique implementation with too early cleanup
• Don't include the visited websites without a nice title in message log

[3.0.3] - 2025-10-13

• Fix potential error in open website logic if response not defined
• Better token limit handeling
• Internal knowledge base page referencing

[3.0.2] - 2025-10-10

• Get website title for OpenAI agent
• Bolding of message logs
• Clarifying questions and research brief dependent on engine type

[3.0.1] - 2025-10-08

• Improved citation logic supporting internal search documents
• Fixed bug in referencing of internal sources not giving the correct title of sources
• OpenAI engine converted to async processing to not be blocking
• Prompt improvements
• Small changes to message logs
• Improve success rate of website title extraction
• Web_fetch tool improvements on error handeling for llm

[3.0.0] - 2025-10-07

• Simplification and better descriptions of configuration
• Dynamic tool descriptions and improved prompts
• Reduce OpenAI engine logging

[2.1.3] - 2025-10-06

• Error handeling on context window limits

[2.1.2] - 2025-10-02

• Remove temperature param to allow for more models used in unique custom
• Research prompt improvements
• Citation rendering improvements with extra llm call
• Additional logging for openai and custom agent

[2.1.1] - 2025-10-01

• bugfix of langgraph state issue
• more logging

[2.1.0] - 2025-10-01

Prompt improvements

• Pushing agent for deeper analysis and including tool descriptions

[2.0.0] - 2025-09-26

Simplification, bugfixes, and performance improvements

• Improve lead and research agent prompts
• Simplify configuration of tool
• root try-catch for error handeling
• Prompt engineering on report writer prompt to ensure inline citations
• Simplify thinking messages
• Include url title for web_fetch

[1.1.1] - 2025-09-23

Minor bugfixes:

• Message log entry at the completion of the report
• Improved instruction on followup questions to use numbered list instead of bullets
• Bugfix of internalsearch and internalfetch due to breaking change in toolkit
• Stronger citation requirements in prompt

[1.1.0] - 2025-09-23

• Use streaming for followup questions and only a single iteration allowed
• Set default models to GPT 4o for followup and GPT 4.1 for research brief

[1.0.0] - 2025-09-18

• Bump toolkit version to allow for both patch and minor updates

[0.0.11] - 2025-09-17

• Updated to latest toolkit

[0.0.10] - 2025-09-12

• Upgrade web search version

[0.0.9] - 2025-09-11

• Bugfixes of statemanagement
• missing tool call handlers
• small performance improvements

[0.0.8] - 2025-09-09

• Implement custom deep research logic using langgraph

[0.0.7] - 2025-09-05

• Set message execution to completed
• Better error protection
• Formatting of final output report

[0.0.6] - 2025-09-04

• Fix null pointer issue in web search action query handling

[0.0.5] - 2025-09-04

• Additional messages in message log and add formatted messages in details

[0.0.4] - 2025-09-02

• Introducing handover capability.

[0.0.3] - 2025-09-03

• Bump toolkit version to get bugfix and small cleanup of getting client

[0.0.2] - 2025-09-02

• Update standard config to use existing LMI objects

[0.0.1] - 2025-09-01

• Initial release of deep_research