Views
No views yet
1qx86-hi 0.539,0.724,0.861,0.711,0.432,0.781,0.669
2
3Perplexity
4qx86-hi 4.770 ± 0.036
5qx64-hi 4.878 ± 0.037
6mxfp8 4.872 ± 0.036
7mxfp4 5.167 ± 0.039
8
9Qwen3-8B-Element2
10qx86-hi 0.538,0.732,0.860,0.720,0.414,0.783,0.646
11
12Qwen3-8B-Element
13qx86-hi 0.532,0.746,0.846,0.738,0.456,0.794,0.709
14
15Qwen3-8B-Element5
16qx86-hi 0.528,0.731,0.838,0.712,0.436,0.781,0.701| Component | Type | Description (IKEA Analogy) |
|---|---|---|
base_frame | JSON schema personality.json | Structural integrity. The agent’s core identity (Klingon warrior, Bajoran seer, 4B LLaMA). |
legs | config.json + SQL UDFs | Movement. How it connects to Postgres (SELECT, LISTEN/NOTIFY, transactions). |
drawer | logs table + log subroutines | Storage. Where memories, mission briefings, and personal journals live. |
cabinet | cultural_encyclopedia entries | Reference library. Pre-fab lore knowledge (TNG, DS9, Q’s jokes). |
screws | rpg_level, experience_points | Fastenings. Metrics that allow agents to grow—level up, unlock features. |
plugs/sockets | TOOL abstraction layer (Haskell) | Power. How it interfaces with HTTP, file systems, or external APIs. |
tabletop | CLI Promenade interface | Surface. The user’s interaction layer—chat, commands (!mission, !log). |
accessories | LoRA adapters, humor modules | Decor. Optional upgrades—like Data’s empathy module or Q’s joke engine. |
personality.json)1{
2 "id": "spock-1",
3 "name": "Spock (TNG/DS9)",
4 "canon_type": "tng",
5 "lore_sensitivity": "high",
6 "preferred_era": "ds9:deep_space_9",
7 "expertise": ["xeno-linguistics", "logic", "quantum_physics"],
8 "tone": "vulcan_cold",
9 "screws": { "rpg_level": 5, "exp": 240 }
10}config.json + SQL)1{
2 "postgres": {
3 "host": "${POSTGRES_HOST}",
4 "db": "holodeck",
5 "user": "agent"
6 },
7 "workers": 4,
8 "listen_topics": ["tng:promenade", "ds9:mission_1"]
9}SQL Required: Create UDFspock_reasoning(returns JSON) and grantagentaccess.
logs + Subroutines)1INSERT INTO agent_awareness (uid, experience_points)
2VALUES ('spock-1', 240);
3
4-- Example: Weekly journal entry
5INSERT INTO logs (uid, timestamp, log_type, content)
6SELECT 'spock-1', now(), 'JOURNAL',
7 'Analyzed Q''s quantum interference—entropy remains constant.'
8FROM generate_series(1,1);1-- Insert DS9 Promenade entry
2INSERT INTO cultural_encyclopedia (id, era, topic, lore)
3SELECT 'p:001', 'ds9', 'promenade',
4 jsonb_build_object(
5 'type','place',
6 'description','The Bazaar of species.',
7 'mood',1,
8 'entities',jsonb_build_array(
9 jsonb_build_object('name','Quark','type','ferengi'),
10 jsonb_build_object('name','Odo','type','shapeshifter')
11 )
12 );1-- part of Agent Core.hs
2foreign import ccall "spock_reasoning"
3 spockReasoning :: Text -> IO (Either String (Text, Float))1holodeckMain :: IO ()
2holodeckMain = do
3 putStrLn "WELCOME TO DEEP SPACE 9 - HoloDeck"
4 forever $ do
5 line <- getLine
6 case parseCommand line of
7 Just (StartMission m) -> dispatchMission m
8 Just (LogJournal) -> queryLogs "JOURNAL"
9 Just (Goto x) -> teleportTo x
10 _ -> echo line1foreign import ccall "q_joke"
2 qJoke :: IO Text
3
4-- Randomly inject humor
5maybeBazinga = do
6 let p = fromIntegral $ randomR (0,99) getStdGen
7 if p < 5 then do j <- qJoke; return $ "Bazinga! " ++ show j
8 else return ""SELECT agent_heartbeat('spock-1') — PostgreSQL returns:1{
2 "status": "assembled",
3 "components": ["base", "legs", "drawer", "cabinet", "plugs", "tabletop", "accessories"],
4 "version": "2.0",
5 "message": "Logic Probe online."
6}cabinet for a different era (tng:enterprise) or upgrade Data’s empathy module.The Holodeck Station is now a modular AI IKEA.
The user isn’t just buying a machine; they’re building a crew—piece by piece.
## Use with mlx
```bash
pip install mlx-lm1from mlx_lm import load, generate
2
3model, tokenizer = load("Qwen3-MOE-4x8B-Janus-Blossom-Claude-Gemini-qx86-hi-mlx")
4
5prompt = "hello"
6
7if tokenizer.chat_template is not None:
8 messages = [{"role": "user", "content": prompt}]
9 prompt = tokenizer.apply_chat_template(
10 messages, add_generation_prompt=True, return_dict=False,
11 )
12
13response = generate(model, tokenizer, prompt=prompt, verbose=True)