adj-ai
adj-ai is a multi-provider LLM orchestrator. It generates several candidate
responses to a task from one or more model providers (Anthropic, OpenAI,
DeepSeek), then uses a "judge" model to adjudicate the candidates and return the
best one.
The core idea is analyse-delegate-judge: by producing multiple candidates and scoring them against a judging rubric, you get higher-quality outputs than a single one-shot completion.
Status: Early development. The Anthropic, OpenAI, and DeepSeek generation/judging paths are all implemented.
How it works
- Model selection – A lightweight model (e.g. Anthropic Haiku) inspects the task prompt and picks the most appropriate generation model, balancing cost and performance.
- Candidate generation – The chosen model produces
candidatesresponses to thetask_prompt. - Judging – A stronger model (e.g. Anthropic Sonnet) ranks the candidates
using the
judge_promptand returns them in order of preference. - Result – The top-ranked candidate is returned.
When multiple clients are configured, judging falls back in priority order: Anthropic → OpenAI → DeepSeek.
Installation
pip install adj-ai
This pulls in the official provider SDKs (anthropic, openai) automatically.
DeepSeek and OpenAI share the
openaiSDK (DeepSeek exposes an OpenAI-compatible API).
Usage
from anthropic import Anthropic
from openai import OpenAI
from adj_ai import Orchestrator
anthropic_client = Anthropic(api_key="...")
openai_client = OpenAI(api_key="...")
deepseek_client = OpenAI(api_key="...", base_url="https://api.deepseek.com")
orch = Orchestrator(
task_prompt="Write a concise product description for a smart water bottle.",
judge_prompt="Pick the description that is clearest and most persuasive.",
anthropic_client=anthropic_client,
openai_client=openai_client,
deepseek_client=deepseek_client,
candidate_number=5,
max_task_tokens=1000,
)
best = orch.run_task()
print(best)
At least one of anthropic_client, openai_client, or deepseek_client must be
provided, otherwise an OrchestratorError is raised.
Configuration
| Parameter | Type | Default | Description |
|---|---|---|---|
task_prompt |
str |
— | The task to generate candidate responses for. |
judge_prompt |
str |
— | The rubric used to rank candidates. |
anthropic_client |
Anthropic |
— | Optional Anthropic client. |
openai_client |
OpenAI |
— | Optional OpenAI client. |
deepseek_client |
OpenAI |
— | Optional DeepSeek (OpenAI-compatible) client. |
candidate_number |
int |
1 |
Number of candidate responses to generate. |
max_task_tokens |
int |
1000 |
Maximum tokens per generated candidate. |
TODO
License
Released under the MIT License.
Metadata
Release files for adj-ai 1.0.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| adj_ai-1.0.1.tar.gz | 8.6 kB | With provenance | Details for adj_ai-1.0.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| adj_ai-1.0.1-py3-none-any.whl | Python 3 | none | any | Details for adj_ai-1.0.1-py3-none-any.whl |
Total release size: 16.1 kB
Release files / adj_ai-1.0.1.tar.gz
| Download URL | adj_ai-1.0.1.tar.gz |
|---|---|
| Size | 8.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
305a0082233796e0624f8e0843e3f4044432c38db19a4af143380986c55dfae2
|
|
BLAKE2b-256 checksum How to use checksums |
9da0064d49354d1ab8168662b045f228101084c073d7d9acef8b7f0f694b506b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jul 2, 2026.
Transparency logRelease files / adj_ai-1.0.1-py3-none-any.whl
| Download URL | adj_ai-1.0.1-py3-none-any.whl |
|---|---|
| Size | 7.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c3e1b6270610ccdc4b5e125e5386713b834dff90583aa8f50e1ac86a52726d29
|
|
BLAKE2b-256 checksum How to use checksums |
90cb8e460b4a2941ea4aedc65161081a1a527a74e2e22e1b31f2fb52eed8246d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jul 2, 2026.
Transparency logSecurity for adj-ai 1.0.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for adj_ai-1.0.1.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jul 2, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.