Skip to main content

Deep, recursive, goal-driven LLM explorer

Project description

deepllm: Full Automation of Goal-driven LLM Dialog Threads with And-Or Recursors and Refiner Oracles



We automate deep step-by step reasoning in an LLM dialog thread by recursively exploring alternatives (OR-nodes) and expanding details (AND-nodes) up to a given depth. Starting from a single succinct task-specific initiator we steer the automated dialog thread to stay focussed on the task by synthesizing a prompt that summarizes the depth-first steps taken so far.

Our algorithm is derived from a simple recursive descent implementation of a Horn Clause interpreter, except that we accommodate our logic engine to fit the natural language reasoning patterns LLMs have been trained on. Semantic similarity to ground-truth facts or oracle advice from another LLM instance is used to restrict the search space and validate the traces of justification steps returned as answers. At the end, the unique minimal model of a generated Horn Clause program collects the results of the reasoning process.

As applications, we sketch implementations of consequence predictions, causal explanations, recommendation systems and topic-focussed exploration of scientific literature.


First, you will need to acquire your OpenAI key from here.

NEW: with it, you are ready to try out at:

NEW: an intro on how to use the app and the API is now on Youtube

To run the code locally, put the OpenAI key in your Linux or OS X shell environment with:

export OPENAI_API_KEY=<your_key>


Clone from github with with:

git clone


If you have cloned this repo, you can install the package deepllm by typing in folder recursors

pip3 install -e .

You can also install it from pypi with

pip3 install deepllm


The DeepLLM API exposes its high-level functions ready to embed in your application with something as simple as (assuming the your OPENAI_KEY is exported by your environment):

for result in run_recursor(initiator='Using tactical nukes', prompter=conseq_prompter, lim=2):

Also, you can explore questions with less gruesome results like in:

for result in run_rater(initiator='Artificial General Intelligence', prompter=sci_prompter, lim=2, threshold=0.5):

Tests and demos

Streamlit web app

After installing streamlit, try it with:

streamlit run deepllm/apps/

Paper describing this work

If you find this software useful please cite it as:

       author = {{Tarau}, Paul},
        title = "{Full Automation of Goal-driven LLM Dialog Threads with And-Or Recursors and Refiner Oracles}",
      journal = {arXiv e-prints},
     keywords = {Computer Science - Artificial Intelligence, Computer Science - Logic in Computer Science},
         year = 2023,
        month = jun,
          eid = {arXiv:2306.14077},
        pages = {arXiv:2306.14077},
          doi = {10.48550/arXiv.2306.14077},
archivePrefix = {arXiv},
       eprint = {2306.14077},
 primaryClass = {cs.AI},
       adsurl = {},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}

You can also find the paper (and future related work) in folder docs.


Paul Tarau

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

deepllm-1.0.3.tar.gz (44.8 kB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page