AGENT-K Python Backend
Multi-LLM Council Backend for AGENT-K
╭──────────────────────────────────────────────── AGENT-K v2.3.7 ────────────────────────────────────────────────╮
│ Multi-LLM Council - GPT + Gemini + Claude │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
The Python backend provides multi-LLM consensus and Smart Context Selection for AGENT-K.
Features
- Council Mode - Three-stage consensus with GPT-4, Gemini, and Claude
- Smart Context Selection - RLM-inspired file selection using LLM reasoning
- Scout Agent - Intelligent codebase research with query-aware file selection
- LiteLLM Integration - Unified API for multiple LLM providers
Installation
pip install agentk8
Requirements
- Python 3.10+
- API Keys (set as environment variables):
OPENAI_API_KEY- For GPT-4GEMINI_API_KEY- For GeminiANTHROPIC_API_KEY- For Claude
Quick Start
Scout Agent (Smart Context Selection)
from agentk.scout import Scout
import asyncio
async def main():
scout = Scout(project_root="/path/to/project")
# Query-aware file selection
context = await scout.scan_project("Where is authentication handled?")
print(context["files"]) # Returns only relevant files
# Full investigation with web search
report = await scout.investigate("Latest JWT best practices")
print(report.to_context_string())
asyncio.run(main())
Council Mode (Multi-LLM Consensus)
from agentk.council import Council
import asyncio
async def main():
council = Council()
# Three-stage consensus
result = await council.deliberate(
"Design a rate limiting system for our API",
mode="council" # or "solo" for multi-Claude personas
)
print(result["final_synthesis"])
asyncio.run(main())
Council Architecture
┌─────────────────────────────────────┐
│ Stage 1: Analysis │
│ GPT-4 | Gemini | Claude (parallel) │
└─────────────────┬───────────────────┘
│
┌─────────────────▼───────────────────┐
│ Stage 2: Cross-Review │
│ Each model reviews others' work │
└─────────────────┬───────────────────┘
│
┌─────────────────▼───────────────────┐
│ Stage 3: Chairman Synthesis │
│ Claude synthesizes consensus │
└─────────────────────────────────────┘
Smart Context Selection (RLM-Inspired)
Instead of blindly grabbing files, the Scout asks the LLM to select relevant files:
# Traditional approach (naive):
files = get_top_10_files() # Often irrelevant
# Smart Context Selection:
scout = Scout(project_root=".")
context = await scout.scan_project("How does the auth middleware work?")
# LLM analyzes file tree + query → selects only auth-related files
This is inspired by the Recursive Language Models paper, which treats the codebase as an environment to navigate intelligently.
CLI Usage
# Run Scout investigation
python -m agentk.scout "Where are the API endpoints defined?"
# Run Council deliberation
python -m agentk.council "Design a caching strategy" --mode council
Module Structure
agentk/
├── __init__.py
├── council.py # Multi-LLM consensus logic
├── scout.py # Smart Context Selection
├── llm.py # LiteLLM wrapper for unified API
└── tools.py # File tree, directory scanning
Environment Variables
| Variable | Description |
|---|---|
OPENAI_API_KEY |
OpenAI API key for GPT-4 |
GEMINI_API_KEY |
Google API key for Gemini |
ANTHROPIC_API_KEY |
Anthropic API key for Claude |
License
MIT License
AGENT-K v2.3.7 - Python Backend
Metadata
Release files for agentk8 2.3.8
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 |
|---|---|---|---|---|
| agentk8-2.3.8.tar.gz | 19.7 kB | Without provenance | Details for agentk8-2.3.8.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agentk8-2.3.8-py3-none-any.whl | Python 3 | none | any | Details for agentk8-2.3.8-py3-none-any.whl |
Total release size: 40.5 kB
Release files / agentk8-2.3.8.tar.gz
| Download URL | agentk8-2.3.8.tar.gz |
|---|---|
| Size | 19.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
40ab658e3636e57a54901a40c9f9ec615b967b1cd84d431ba5133b64c99f63e0
|
|
BLAKE2b-256 checksum How to use checksums |
b995118ff351d42053a1edc44a5e030ed479bbb4cb820a8d9ec477780330690e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Release files / agentk8-2.3.8-py3-none-any.whl
| Download URL | agentk8-2.3.8-py3-none-any.whl |
|---|---|
| Size | 20.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
e944d45437383e21694877c3b90525e86b65bfd96e5b9298b13aef42f87cbae6
|
|
BLAKE2b-256 checksum How to use checksums |
18426e22febfd18783e8cea77310192d5997f1ead1993b84e33fe5e21bbcf98e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Security for agentk8 2.3.8
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.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A 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.