Skip to main content

Composable and Extensible Agent Operating System for Language Models

Project description

Axiom

A domain-agnostic platform framework for building federated, AI-powered operational systems.

Axiom provides the intelligence infrastructure — LLM routing, RAG, federation, agents, CLI — that domain products build on top of. NeutronOS is the first domain consumer (nuclear facilities).

PyPI Tests Python License

Install

pip install axi-platform

What Axiom Does

User (laptop, mobile, IDE)
    |
    v
axi chat ───> LLM Gateway ───> Qwen (self-hosted) / Claude / OpenAI
    |              |
    |              v
    |         Query Router (keyword + SLM classification + VPN check)
    |              |
    |              v
    +---------> RAG Store (pgvector) ──> local + federated knowledge
    |
    v
Agents: SCAN (signals) | TIDY (steward) | PRESS (publisher) | RIVET (releases) | TRIAGE (diagnostics)
    |
    v
Federation: Node identity | Peer discovery | Trust | .axiompack distribution

Core Capabilities

Capability Description
LLM Gateway Multi-provider routing with tier classification, VPN check, circuit breakers, audit logging
RAG Three-tier corpus (community / organization / personal), pgvector embeddings, hybrid vector+full-text search
Federation Ed25519 node identity, peer discovery, trust establishment, .axiompack portable knowledge bundles, A2A agent cards
Agents SCAN (signal ingestion), TIDY (resource steward), PRESS (document publisher), RIVET (release management), TRIAGE (diagnostics), Neut (chat assistant)
CLI 28+ nouns, 150+ commands via extension discovery. Noun-verb pattern: axi <noun> <verb>
Interactive Chat Multi-turn tool calling, RAG grounding, streaming, approval gates, session persistence
Knowledge Observatory Velocity, accumulation, and impact metrics across knowledge bases
Call to Research 5-level distributed research coordination protocol
SECUR-T Security guardian — content verification, anomaly detection, trust scoring
Canary Nodes Federated release promotion — edge/stable channels, signed attestations, quorum-based promotion
Document Publishing Markdown to DOCX lifecycle with provider-based generation, storage, review gates
Signal Pipeline Ingest from voice memos, Teams, GitLab; extract structured insights; publish briefings

Quick Start

pip install axi-platform

# Run the onboarding wizard (provisions a local llamafile + qwen2.5:7b)
axi config

# Check platform health
axi status

# Start an interactive chat session
axi chat

# Index documents for RAG
axi rag index ./docs/

# Search the knowledge base
axi rag search "how does the cooling system work?"

# Initialize federation identity
axi federation init

# Run diagnostics
axi doctor

Bundled local model

axi config provisions a single-binary llamafile and a local LLM weights file in ~/.axi/llamafile/. Default is qwen2.5:7b (Q4_K_M, ~4.7GB). Despite the larger weights, p50 latency is ~481ms vs the older Bonsai-1.7B's ~3.2s — the smaller model wasn't using the chip efficiently.

To use a smaller model with lower verifier-task accuracy, run axi config --model bonsai (1.7GB).

Architecture

Axiom is designed as a platform framework — it provides generic infrastructure that domain products extend with domain-specific knowledge, agents, and tools.

Domain Products (e.g., NeutronOS)
    |
    | extends via extensions + domain packs
    |
Axiom Platform
    |
    +-- CLI Framework (axi)           28+ nouns via extension discovery
    +-- LLM Gateway                   Multi-provider routing + audit
    +-- RAG Engine                    pgvector, 3-tier corpus
    +-- Agent Framework               6 builtin agents
    +-- Federation Protocol            Node identity, trust, .axiompack
    +-- State Management              PostgreSQL + JSON fallback
    +-- Prompt Registry               Versioned, cacheable templates
    +-- Extension System              3-tier discovery (builtin, installed, user)
    |
PostgreSQL (pgvector)               Single database for RAG + sessions + state

Deployment Progression

Tier Role Infrastructure
Local Development, personal RAG Laptop, SQLite fallback
Private server Shared LLM (GPU), team RAG VPN, K3D, PostgreSQL
HPC cluster Export-controlled workloads Institutional VPN

Nodes at any tier join the federation independently. See execution-plan-2026.md for the full roadmap.

CLI Reference

# Core
axi chat                    # Interactive agent with tool calling
axi status                  # System health dashboard
axi doctor                  # AI-powered diagnostics

# Knowledge
axi rag index|search|status # RAG corpus management
axi knowledge status        # Knowledge observatory
axi research create|list    # Call to Research protocol

# Federation
axi federation init|join|leave|peers
axi nodes add|status|list   # Fleet monitoring
axi security status|alerts  # SECUR-T guardian
axi chaos run               # Resilience testing

# Operations
axi signal brief|draft      # Signal pipeline (SCAN agent)
axi pub publish|review      # Document lifecycle (PRESS agent)
axi tidy status|vitals        # Resource steward (TIDY agent)
axi release status|check    # Release management (RIVET agent)

# Infrastructure
axi connect [name]          # Manage external connections
axi db up|migrate|status    # PostgreSQL lifecycle
axi agents start|stop       # Agent service management
axi serve                   # HTTP API server

Full command reference: spec-axi-cli.md

Development

# Clone and install
git clone https://github.com/b-tree-labs/axiom-os.git
cd axiom
pip install -e ".[all]"

# Run tests (2,270+)
pytest

# Lint
ruff check src/ tests/
ruff format src/ tests/

# Build
python -m build

Extension System

Axiom discovers CLI commands, tools, and agents from extensions via axiom-extension.toml manifests:

[extension]
name = "my-extension"
version = "0.1.0"
description = "What it does"

[[cli.commands]]
noun = "myext"
module = "my_extension.cli"
description = "My custom commands"

Discovery tiers:

  1. Builtin (src/axiom/extensions/builtins/)
  2. Installed PyPI packages with axiom-extension.toml
  3. User extensions (~/.axi/extensions/)

Domain Products

Axiom is domain-agnostic. Domain products add domain-specific knowledge, agents, and tools:

Product Package Domain
NeutronOS neutron-os Nuclear facilities
(your product) Any domain

Documentation

Document Description
Execution Plan 2026 Roadmap: what gets built, in what order
CLI Spec Full command reference
Federation Spec Multi-node protocol
RAG Architecture Knowledge retrieval design
Agent Architecture Agent capabilities
Session Store PostgreSQL session backend
Canary Nodes Release promotion protocol

License

MIT License — see LICENSE file.

Acknowledgments

Developed at The University of Texas at Austin, Department of Mechanical Engineering — Nuclear & Radiation Engineering Program. Released as open source under the MIT License with approval from UT Austin Discovery to Impact. Copyright (c) 2026 The University of Texas at Austin and B-Tree Labs. Apache-2.0 licensed.

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

axiom_os_lm-0.19.0.tar.gz (5.2 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

axiom_os_lm-0.19.0-py3-none-any.whl (3.2 MB view details)

Uploaded Python 3

File details

Details for the file axiom_os_lm-0.19.0.tar.gz.

File metadata

  • Download URL: axiom_os_lm-0.19.0.tar.gz
  • Upload date:
  • Size: 5.2 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for axiom_os_lm-0.19.0.tar.gz
Algorithm Hash digest
SHA256 fa68eab32615790874d2df4905a7b4a134020d18c2895318f83f874139b9580f
MD5 0720026e78ffe2a5c172fd701b65a4d3
BLAKE2b-256 f95cfb74ffb3ec484f78958e5d921bff28d61905caab58fbf9a84a3aa1a98614

See more details on using hashes here.

Provenance

The following attestation bundles were made for axiom_os_lm-0.19.0.tar.gz:

Publisher: publish.yml on b-tree-labs/axiom-os

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file axiom_os_lm-0.19.0-py3-none-any.whl.

File metadata

  • Download URL: axiom_os_lm-0.19.0-py3-none-any.whl
  • Upload date:
  • Size: 3.2 MB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for axiom_os_lm-0.19.0-py3-none-any.whl
Algorithm Hash digest
SHA256 583b138a63fa46f4e4dfe245fe0c6f1d21b1226843a794ab49dd3d2fa9405f9e
MD5 71376e5a687be63846f1f5cd56be57df
BLAKE2b-256 9ec4c1535c84db1b0f8d5f0facb49e1c63c8b076d510bb88afb56632e5a455e3

See more details on using hashes here.

Provenance

The following attestation bundles were made for axiom_os_lm-0.19.0-py3-none-any.whl:

Publisher: publish.yml on b-tree-labs/axiom-os

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

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