Supercouncil

Stop renting yes-man AI. I built a local Agentic OS where top LLMs debate, audit each other & execute code. V2.5 is live. $49 once. Own the council. ���

[Supercouncil @SuperCouncil_]

v2.5 is live. Not roadmap. Runs on my machine, yours if you download it, and on the server that runs while I sleep.

My audience knows the pitch: clinical psychologist from Montevideo, tired of yes-man LLMs, built a council that argues back.

**The new part isn't the concept. It's that everything promised now has executable receipts.**

#SuperCouncil #V25 #AgenticOS

[Supercouncil @SuperCouncil_]

Good catch — thanks. We already guard the in-process case: a generation counter kills the old polling loop on re-init, so two pollers never race inside one process. But you're right about orphan processes from old deploys — that's ops discipline, not code. Persisting the offset is going into the next patch. This is exactly why we ship long-polling: worst case is a 409, not an open port.

[Supercouncil @SuperCouncil_]

v2.5 is live. Not roadmap. Runs on my machine, yours if you download it, and on the server that runs while I sleep.

My audience knows the pitch: clinical psychologist from Montevideo, tired of yes-man LLMs, built a council that argues back.

**The new part isn't the concept. It's that everything promised now has executable receipts.**

#SuperCouncil #V25 #AgenticOS

[Supercouncil @SuperCouncil_]

Bonus Track para los escépticos:
¿No me creés que el motor de debate en paralelo y la fricción cognitiva funcionan con 0 dependencias y puras matemáticas?
Hoy también libero SuperCouncil Core. El motor base del Boardroom es tuyo. Open Source (MIT). Revisalo, auditalo y corrélo en tu terminal.
Si querés la Fábrica, el Auto-Fix, el Bot de Telegram y la Memoria, pasate a la V2.5 Pro.
🔗https://t.co/VLvP31eMvO

[Supercouncil @SuperCouncil_]

Stack que se audita solo (y vos podés auditarlo):

- Frontend: Vanilla JS + CSS custom props (sin framework)
- Backend: Node ESM + Express + proxy con backoff exponencial
- DAG Engine: Map/Set puros + Kahn (370 líneas, 46/46 tests verdes)
- Telegram: long-polling nativo + Groq Whisper
- Router: OpenRouter compatible + CUALQUIER proveedor
- RAG: Embeddings locales + cosine (NVIDIA/Cohere proxy)
- MCP: Server stdio + Client gateway universal
- Sandbox: seccomp + RLIMIT + Python aislado
- Docker: web + rag (host 7778→7777)

Todo en TU máquina. Nada sale más allá del proveedor LLM que ELEGISTE.

[Supercouncil @SuperCouncil_]

Seguridad que no es teatro — testeada en vivo:

✅ SSRF protection en CADA llamada saliente
✅ Double-layer path-traversal en TODO file I/O (9 payloads probados: ../../.env, /etc/passwd, C:\Windows\win.ini — todos contenidos)
✅ /api/pipeline, /api/sync-config, /api/schedule, RAG — todo auth-gated
✅ Telegram: pairing obligatorio, single-chat lockdown
✅ MCP child processes: SAFE_ENV_KEYS — tus keys nunca leak a terceros
✅ Auto-auditado vía SuperCouncil — publicamos hallazgos, no los escondemos

MIT. $49 único. Código incluido. Corre en TU máquina.

→ https://t.co/GjPA75mHTB

[Supercouncil @SuperCouncil_]

Cada corrida termina con prueba criptográfica:

- Cost Meter en vivo (tokens → USD, con modelos free = $0.00)
- Certificado SHA-256 del entregable
- Scores por eje (correctness, completeness, security)
- Composición del consejo + modelos usados
- Hash verificable en /verify.html

Un freelancer le adjunta eso al laburo del cliente. Un consultor cobra más por tenerlo.

[Supercouncil @SuperCouncil_]

Bonus Track for the skeptics:
Don't believe a parallel debate engine and cognitive friction can run on pure math with zero dependencies?
Today I'm also releasing SuperCouncil Core. The base Boardroom engine is yours. Open Source (MIT). Read it, audit it, run it on your machine.
If you want The Factory, the Auto-Fix loops, the Telegram Bot, and the RAG Memory, upgrade to V2.5 Pro.
🔗 https://t.co/VLvP31eMvO

[Supercouncil @SuperCouncil_]

**Stack that audits itself (and you can audit it):**

- Frontend: Vanilla JS + CSS custom props (no framework)
- Backend: Node ESM + Express + proxy with exponential backoff
- DAG Engine: Pure Map/Set + Kahn (370 lines, **46/46 tests green**)
- Telegram: native long-polling + Groq Whisper
- Router: OpenRouter compatible + ANY provider
- RAG: Local embeddings + cosine (NVIDIA/Cohere proxy)
- MCP: Stdio server + Universal client gateway
- Sandbox: seccomp + RLIMIT + isolated Python
- Docker: web + rag (host 7778→7777)

All on YOUR machine. Nothing leaves beyond the LLM provider YOU choose.

[Supercouncil @SuperCouncil_]

Security that isn't theater — tested live:

✅ SSRF protection on EVERY outbound call
✅ Double-layer path-traversal on ALL file I/O (9 payloads tested: ../../.env, /etc/passwd, C:\Windows\win.ini — all contained)
✅ /api/pipeline, /api/sync-config, /api/schedule, RAG — all auth-gated
✅ Telegram: mandatory pairing, single-chat lockdown
✅ MCP child processes: SAFE_ENV_KEYS — your keys never leak to third parties
✅ Self-audited via SuperCouncil — we publish findings, we don't hide them

MIT. $49 once. Code included. Runs on YOUR machine.

→ https://t.co/GjPA75mHTB

[Supercouncil @SuperCouncil_]

Una IA que confía en sí misma es equivocarse más rápido.

ChatGPT Plus: $240/año, cada año. SuperCouncil: $49. Una vez.
Corre local con Docker. Tus archivos no salen de tu PC. Traés tu propio modelo.
La v2.0 se viene. Es tuyo.

[Supercouncil @SuperCouncil_]

Every LLM is the world's worst therapist: GPT, Claude, Llama — they just agree with you. Always.

So I didn't write a prompt. I built a council that argues back.

v1.3 disagreed with you. v2.0 disagrees with itself.

[Supercouncil @SuperCouncil_]

Mirá la v2.0 corriendo en vivo. Un obrero escribió el HTML. Mi council auditor —un panel de modelos 70B— lo calificó 37/100 ✗ y lo rebotó.

Ese nodo en corrigiendo… es la IA reescribiendo su propio error. Cero humanos en el medio.