ain-state-compiler 0.1.0

The G-Brain Company Brain Primitive: continuously compiles Slack, Jira, and Gmail into an executable, conflict-resolved operational state for AI agents.

ain-state-compiler

The G-Brain Primitive — A local-first, self-improving Company Brain that continuously compiles Slack, Jira, and Gmail into an executable, conflict-resolved operational state for AI agents.

---

What Is This?

Most enterprise knowledge systems offer static storage and RAG over documents.

ain-state-compiler solves a harder problem:

Continuously compiling an organization's operational reality into an executable, internally consistent state representation.

The output is not a document retrieval system — it is a Company Brain:

Dimension	Static KB (Notion/Wikis)	AIN State Compiler
Focus	Storing documents	Maintaining live state
Refunds	Refund policy PDF	Actual refund behaviors observed across support logs
Engineering	Static runbooks	SRE behavior during actual incident response
Workflows	SOP text	Extracted from real event execution traces

---

Architecture

[ Slack / Jira / Gmail ]
         |
         v
[ Shared Hivemind DB ]   <-- Supabase (cloud) or SQLite (local)
         |
         v  (Every 5 minutes, offline cron)
[ sync + state_compiler ]  -- Zero-LLM, 100% offline
         |
    +---------+-----------+
    |         |           |
[IMMs]   [OEGs]    [Conflict Report]
    \         |           /
         [compiled_state/]
              |
    (On-demand query only)
              v
        [ Ollama LLM ]   <-- local gemma3:1b
              |
        [ Agent Action ]

Key Principles:

• Zero-LLM at source: All ingestion, structuring, and conflict detection is 100% offline deterministic code.
• LLMs on-demand only: Only invoked when you explicitly query the brain. Context is pre-compiled and token-optimized (14-30% reduction via JSON→YAML).
• Self-improving: Notices recurring resolution structures and updates its executable skill library.
• Shared hivemind: Multiple team members write to one central DB; all local nodes compile their own state and read from the same source of truth.

---

Installation

pip install ain-state-compiler

First-Time Setup

After installing, run the interactive setup wizard:

ain-brain setup

This will prompt you for:

• Slack Bot Token (xoxb-...)
• Jira URL + API Token
• Gmail App Password
• Supabase URL + Service Role Key (central cloud hivemind)
• Optional external DB credentials

All credentials are saved to .env in your project directory.

Install Ollama (Local LLM)

ain-brain install-ollama

This downloads and installs Ollama and pulls gemma3:1b. LLMs are only called when you query the brain — never during ingestion.

Register 5-Minute Background Sync

ain-brain cron

• Windows: Registers a Windows Task Scheduler task (AINBrainSync).
• Linux/macOS: Adds a crontab entry.

The background task runs sync + compile every 5 minutes completely offline.

---

Usage

Seed the local database (mock data for testing)

ain-brain init-db

Run a one-shot sync + compile

ain-brain sync

Query the Company Brain

ain-brain query "What is the current deployment status of Analytics v2?"
ain-brain query "What is the billing situation with Acme Corp?"
ain-brain query "Are there any active conflicts in the system?"

If Ollama is running, it uses gemma3:1b for inference. Otherwise it falls back to the deterministic offline resolver.

Start the Dashboard Server

ain-brain server

Open http://localhost:8000 to access the B2B glassmorphic dashboard with:

• Live ingest stream terminal
• Active contradictions panel
• YAML OEG viewer
• Interactive query terminal

---

Python API

from ain_state_compiler import StateCompiler, ConflictDetector, TokenOptimizer

# Compile operational state
compiler = StateCompiler("/path/to/project")
summary = compiler.compile()
print(summary)
# {'processed_slack_events': 6, 'processed_jira_issues': 2, ..., 'detected_conflicts': 2}

# Token optimization
import json
data = {"key": "value", "nested": {"a": 1, "b": "hello world"}}
savings = TokenOptimizer.calculate_savings(data)
print(f"YAML saves {savings['saving_percentage']}% tokens vs JSON")

# Query the brain
from ain_state_compiler.query import query_brain
answer, node, used_llm = query_brain("What is the analytics v2 flag status?")
print(answer)

Sync with Supabase

Set environment variables (or configure via ain-brain setup):

SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key

Then run:

ain-brain sync

The sync will pull from Supabase and write compiled state back to the compile_log table.

---

Supabase Schema

Run these SQL commands in Supabase SQL Editor to create the required tables:

CREATE TABLE slack_history (
    id SERIAL PRIMARY KEY,
    channel TEXT NOT NULL,
    timestamp TEXT NOT NULL,
    "user" TEXT NOT NULL,
    text TEXT NOT NULL
);

CREATE TABLE jira_issues (
    id TEXT PRIMARY KEY,
    title TEXT NOT NULL,
    status TEXT NOT NULL,
    assignee TEXT NOT NULL,
    updated_at TEXT NOT NULL,
    description TEXT NOT NULL
);

CREATE TABLE emails (
    id TEXT PRIMARY KEY,
    subject TEXT NOT NULL,
    sender TEXT NOT NULL,
    timestamp TEXT NOT NULL,
    body TEXT NOT NULL
);

CREATE TABLE compile_log (
    id SERIAL PRIMARY KEY,
    compiled_at TEXT NOT NULL,
    slack_events INTEGER,
    jira_issues INTEGER,
    emails INTEGER,
    conflicts INTEGER
);

---

What Gets Compiled

Each sync cycle produces:

File	Description
compiled_state/product_deployment_imm.md	Product Deployment Institutional Memory Module
compiled_state/acme_corp_billing_imm.md	Billing configuration IMM
compiled_state/active_conflicts_report.md	All detected cross-departmental contradictions
compiled_state/operational_state.json	Full operational state (JSON)
compiled_state/operational_state.yaml	Token-optimized YAML (14-30% smaller)
compiled_state/token_optimization_metrics.json	Compression metrics

---

The Contradiction Engine

The primary competitive moat. Instead of exposing raw conflicting data to AI agents (which causes hallucination), the Contradiction Engine resolves state before agents ever see it.

Example detection:

[CON-001] CRITICAL: Feature Flag Rollback vs GA Announcement
  - Jira ENG-1043: Status=Done (GA implied)
  - Email: Marketing sent "Analytics v2 is now live!" to all customers
  - Slack: SRE set analytics_v2=FALSE due to DB connection pool exhaustion

Resolution: Halt GA messaging. Update Jira to BLOCKED. Issue customer notice.

---

Token Optimization

Before submitting context to a local LLM, the JSON→YAML optimizer strips brackets, quotes, and separators:

JSON: {"key": "value", "nested": {"a": 1}}   (38 chars)
YAML: key: value\nnested:\n  a: 1            (27 chars = 29% smaller)

On realistic payloads, this saves 14-30% of tokens, reducing inference latency and API costs.

---

License

MIT License — see LICENSE

---

Built on the AIN (Autonomous Intelligence Network) architecture. Part of the G-Brain Company Brain research program.