Skip to content

Configuration and Variables

Jump to bottom
Brian Vasquez edited this page May 20, 2026 · 5 revisions

Configuration & Variables Overview

Almost every behavior in A.I.M. is driven by a configurable variable exposed in the TUI (aim tui). A.I.M. is fundamentally an operating system for your AI, and it gives you complete control over how the intelligence is routed, governed, and constrained.

⚙️ Bring Your Own LLM (Configuring Providers)

A.I.M. is Provider-Agnostic. While Ollama is the default for free, offline local processing, you are not locked into it. The aim tui interface allows you to swap models, providers, and endpoints for both the Conscious Agent and the Subconscious Daemon.

1. LLM & Cognitive Routing (Per-Tier Customization)

Important Note on the "Primary Brain": The interactive chat model you talk to daily is governed exclusively by your global Gemini CLI settings (~/.gemini/settings.json). The "Primary Brain" configured in the aim tui serves as an explicit fallback model for headless background scripts (like aim audit, aim scrape, or crash recoveries) that execute autonomously under the hood.

  • Providers: Local (Ollama/Llama.cpp), Google Gemini API, OpenAI, Anthropic, Custom OAuth.
  • Models: Swappable per role (e.g., gemini-3.1-pro-preview for the Conscious Agent, gemma4:e4b for the Subconscious Daemon/Tier 1).
  • Endpoints & Auth: Custom API URLs and Auth Types (OAuth, API Key, None).
  • Embeddings: Swappable embedding models and endpoints. While nomic-embed-text running locally is the free default, you can completely elevate your RAG performance by swapping to a premium remote API like Gemini Embedding 2.

🛡️ Guardrails & Anti-Drift (Cognitive Mantra)

The Executive Guardrails actively monitor the agent's behavior to prevent context drift and rogue execution.

  • cognitive_mantra.enabled: Toggle the anti-drift system on or off.
  • whisper_interval: How many tool calls before a silent reminder is injected into the agent's context.
  • mantra_interval: How many tool calls before a hard halt and forced recitation of the GitOps rules.

🧠 Execution & Context Limits

A.I.M. manages the context window and lifecycle of the agent using configurable limits.

  • handoff_context_tail: Number of turns preserved in the dead-man's switch snapshot when generating a handoff.
  • auto_rebirth: Whether the agent automatically spawns its successor upon context collapse (Reincarnation).
  • archive_retention_days: How long to keep historical sessions and artifacts.
  • cleanup_mode: Choose between ARCHIVE or DELETE for consumed files.

🌐 The Sovereign Swarm (P2P)

Control how your local node interacts with the global decentralized intelligence network.

  • swarm_enabled: Toggle BitTorrent DataJack tracking to share or download .parquet cartridges.
  • seeding_ratio and max_download_speed: Traffic shaping and bandwidth limits for the swarm.

🎭 Agent Persona & Architecture

You can configure the high-level identity and operational structure of the agent.

  • Operator Profile: Set execution mode, cognitive level, and conciseness to tune how the agent interacts with you.
  • Cognitive Architecture: Switch between Monolithic, Frontline, or Subconscious modes.
  • Obsidian Vault Path: Configure the local path for the two-way Obsidian Bridge sync.

Launch the cockpit anytime by running aim tui to adjust these settings.

☕ Support the continuous evolution of A.I.M.: Buy Me a Coffee

Navigation

Deep Dives & Philosophy

👉 A.I.M. Knowledge Base (Explore the full design history, benchmarks, and architectural maps in our public Obsidian vault).

Clone this wiki locally