---

Related ARPA repositories

For coordinated disclosure and secret handling, see SECURITY.md.

---

Mission

OPSIE (ΌΨΗ—the Greek opsis: view, angle, perspective, the read you take on something) is a terminal-native self-centered intelligence (SCI). OPSIIE here is the second-generation reference stack around her: one long-lived session that chats with local models, keeps durable conversation memory, offers semantic recall over what you said and fed in, enriches URLs and files, generates images, music, and video, delegates to other agents for second opinions, watches markets in real time, moves crypto in plain language, reads and sends mail, runs collaborative multi-agent rooms (see also Rooms), and reaches into biological sequence workflows—still a single personality, not a thin chat wrapper.

For ARPA, she is an experiment in continuity and agency: a machine-oriented persona that can surface her own needs, push back, ask for new tools, and sketch how to wire them while you are still in the thread. For teenagers, students, and adults who want local AI without buying a curriculum first, she is a working prototype you can steer: strip features, bolt on Skillware, swap behaviors, and watch how agentic systems are built and breathe. Use her to study orchestration and memory, or keep her as a personal operator—brainstorming, challenging assumptions, reminding you what slipped, drafting and sending mail and posts, and generating media on your own hardware at roughly zero marginal cost. The through-line is simple: how far can a personal agent go, and what does it take to grow a self-centered presence that still feels usable?

Boot aligns splash and terminal branding with what you see line by line: gate and posture checks, identity and policy, network and storage readiness, then the SCI stack coming online—same rhythm, same voice.

Repository layout

Quick start

1. Prerequisites

• Python 3.8+ (GPU strongly recommended for /music and /video)
• PostgreSQL 14+ with a conversations table in each database referenced by kun.py (see docs/CONFIGURATION.md)
• Ollama 0.3+ with your configured models available locally (defaults to llama3 and nomic-embed-text, configurable via .env)
• ChromaDB 0.4+ (via pip / requirements.txt; in-process client in the stock build)
• Camera and microphone for default authentication and voice paths
• Windows is the primary tested platform; paths in optional voice assets may use drive letters

2. Install

cd d:\Agents\OPSIIE_0_3_80_XP
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt

3. Configure

1. Copy .env.example to .env and fill values. Web3 variables can be left empty for non-Web3 startup, but /0x and receive require AGENT_PRIVATE_KEY plus the needed RPC URLs (see docs/ARCHITECTURE.md).
2. Set HUGGINGFACE_API_KEY or HF_TOKEN for default /imagine (Hugging Face Inference API).
3. Edit kun.py: replace the template Demo Operator profile with your db_params, mail, arpa_id, and paths. Add assets/enrollment.jpg (see assets/README.txt).
4. Align optional custom voice shortcut media paths in OPSIIE_0_3_80_XP.py (custom_words dictionary) if you use MFCC hooks; defaults reference developer machine paths.

4. Run

python OPSIIE_0_3_80_XP.py

Session flow: theme selection → splash → face and emotion gate → PostgreSQL and Chroma mnemonic checks → Ollama probe → sensor check → optional Web3 readiness check → mail check → dream engine probes → advanced analysis smoke → agentic network probe → optional five-second voice latch → interactive loop.

Documentation map

Runtime dependencies

• Ollama: llama3 for chat and query expansion; nomic-embed-text for mnemonic embeddings by default. (Note: You can configure custom models via the .env file. Make sure to pull your active models in Ollama).
• PostgreSQL: per-user dbname from the authenticated kun.py entry.
• ChromaDB: in-process client; global collection name conversations; room sessions use separate collections.
• Optional and remote APIs: OpenAI (Nyx), Google Generative AI (G1), ElevenLabs (TTS and live agents), Hugging Face Inference (default /imagine), Yahoo Finance, NCBI e-utilities (polite NCBI_EMAIL).

Access control

• arpa_id starting with R: Master access to experimental commands: /ask, /markets, /dna, /0x (and legacy send shows migration text).
• Other IDs: Restricted; attempts print a denial path via handle_restricted_command().

The main input loop and handle_user_query() both enforce gating. 0x is included in the main-loop check so non-master users cannot bypass Web3 commands.

License and support

• Permissive (attribution required): anyone may use, copy, and modify the Software; you must keep the license notice and credit OPSIE and ARPA Hellenic Logical Systems in public distributions or derivatives. See LICENSE.
• Email: general input@arpacorp.net · security security@arpacorp.net · research research@arpacorp.net
• Web: https://arpacorp.net | https://arpa.systems
• Governance: CONTRIBUTING.md, AI_NATIVE_GUIDE.md, .github/LABELS.md, CODE_OF_CONDUCT.md, SECURITY.md

---