⚡ Nokido
Building an artificial organism — not just another AI agent
A local-first runtime that lets local models, cloud providers and coding agents work as one organism — routed, remembered, regulated and gated — on your machine, with your hardware, for your data.
LLM providers (20 vendors · 39 routed slots, by use-case) :
Languages: English · Français · Español · 简体中文 · Português · 日本語 · Deutsch · العربية
Full documentation (nothing removed): docs/README_complet.md — translated full versions: docs/i18n/README_complet.<lang>.md
🧭 In one minute
Nokido is the organism layer around the models: one MCP hub that clients reach, while the system — memory, regulation, gates — persists across sessions. The client stays disposable; the organism does not.
- Routing across local and cloud models by use case, with a local fallback and free-tier quota tracking.
- Long-term memory with provenance — FTS5/BM25 + BGE-M3 (1024-d) vector retrieval, fused and reranked.
- Regulation of CPU, RAM, queues and providers — homeostasis, reflexes, circadian cycles.
- Deterministic gates around actions — AST checks, secret scanning, six-ring RBAC, egress control.
- Agent messaging (M2M, swarms, ACP, A2A) and introspection of its own code and state.
Nokido keeps apart what is declared, observed and verified: a port that answers does not prove a model is
loaded, and a probe that cannot look reports ILLISIBLE (unreadable), never “no”.
🧬 Architecture
NOKIDO
DIGITAL ORGANISM
│
┌──────────────────────┼──────────────────────┐
NERVOUS IMMUNE ENDOCRINE
SYSTEM SYSTEM SYSTEM
events / routing firewall / RBAC resource regulation
M2M / protocols membrane / trust quotas / pressure
└──────────────────────┼──────────────────────┘
┌────────────┴────────────┐
MEMORY MUSCLES
RAG / FTS / vectors workers / tools / execution
└────────────┬────────────┘
CENTRAL INTEGRATION
Hub / MCP · ACP / A2A / M2M / SWARM
│
neural substrate: SNN / edge / NPU → future neuromorphic
Every module declares its organ (__FORGE_COLOR__) and the census forge_module_census --check keeps the anatomy
complete in CI. Details: the organism ·
memory and retrieval ·
homeostasis · swarm cognition.
🚀 Quick start
From PyPI — nokido-agent 0.20.16, published after the
install proof on Linux, Windows and macOS. The complete organism, its pinned components
(checksums verified), then the hub:
pip install "nokido-agent[organisme]==0.20.16"
nokido-composants --tout # deno, llama.cpp, models, knowledge pack, local token
nokido-doctor # what this machine has, lacks, or cannot read
nokido-hub # then: curl http://localhost:8766/health
Client only (talks to an existing hub): pip install nokido-agent==0.20.16.
Alpha. The package published on PyPI installs with pip (see the block above), and each release is proven to install on Linux, Windows and macOS before PyPI serves it. The cloned repository remains the reference path for the full organism — services, Deno supervisor, local models — until the one-command full install is delivered.
| Requirement | Measured |
|---|---|
| Python 3.12+ | 78 of the 108 declared services, and the package itself |
| RAM — 24 GB host recommended | measured 2026-10-19 on the reference host: the full organism holds ~12 GB of private memory (hub, services, local embedder, llama-server) and the RAG cold start adds ~2 GB for ~10 min; the client + SSoT profile (MCP hub + shared memory, no supervisor fleet, no local models) measured 2026-10-30 in a clean venv, median of 3: ~100 MB RSS at rest, 297 MB of venv, install 157 s, hub ready in 1.6 s (tools/bench/mesure_profil_client.py, protocol) |
| Disk — ~60 GB free | the RAG memory alone is ~53 GB (embeddings.db, 1.58 M embedded chunks); the package itself needs ~2 GB |
From a clone — the path CI itself uses:
git clone https://github.com/Nokido-labs/nokido.git
cd nokido
python -m venv .venv
.venv\Scripts\activate # Linux / macOS: source .venv/bin/activate
pip install -r requirements.txt
python tools/nokido_doctor.py # what this machine has, lacks, or cannot read
python tools/nokido_hub.py
With Docker (hub + Ollama):
cp Nokido.env.example Nokido.env
docker compose -f docker/nokido/docker-compose.yml --profile core up -d
Then curl http://localhost:8766/health should answer. pyproject.toml declares 17 extras (plus the all and
full bundles). Requirements, Docker / native paths and verification:
installation · verify ·
first use.
🔌 Connect a client
The hub serves MCP on http://localhost:8766/mcp. Supported clients and transports:
| Client | Transport | Client | Transport | |
|---|---|---|---|---|
| Claude Code | MCP Streamable HTTP | Cline | MCP STDIO | |
| Antigravity (agy) | MCP HTTP | Claude Desktop | MCP STDIO (bridge) | |
| Codex CLI | MCP HTTP | claude.ai · ChatGPT web | MCP HTTPS (cloud peer, quarantined) | |
| OpenCode | MCP HTTP | Mistral Vibe · ZCode · Mammouth Code | MCP HTTP |
Each client gets its own agent token from the vault; an anonymous caller stays at ring 3. ACP (server and client) and an A2A Tier-1 surface complete MCP. Setup: 04 — MCP clients · 24 — Cloud peers.
The hub also exposes GET /api/graph/proprioception: imports / imported-by / call-graph per file, computed from the
AST index of its own source.
🔬 Project status
Each line pairs a declared maturity with the evidence in this repository; tools/forge_capability_audit.py
fails CI when a ✅ has no evidence or a paused service is shown as operational. For the living state, ask the body:
nokido-doctor --vivant (from a clone: python tools/nokido_doctor.py --vivant).
DECLARED EVIDENCE IN THIS REPOSITORY
ANATOMY / ORGANISM
Strict anatomical census ✅ achieved `forge_module_census` --check, 0 unclassified
CI architectural gate ✅ achieved `anatomie` gate, blocking since 2026-09-06
M2M memory separation ✅ achieved `forge_db_path`: one switch read by every process
Emergency homeostasis ✅ achieved `NokidoHomeostasis` + `forge_homeostasis_orchestrator`
Sleep / circadian regulation 🟡 partial all 6 phases fire on measured inactivity (`forge_circadian_runner --auto`); 4 of their 10 service targets disabled by policy
COMMUNICATION
MCP ✅ operational `NokidoMCP` hub, enabled by default
ACP 🟡 partial client side operational (`NokidoAcpWorker` drives claude, codex, opencode, agy sessions); server `NokidoAcpWs` disabled by default
A2A Tier-1 ⏸️ paused `NokidoA2A`, disabled by default (code present)
M2M ✅ operational `forge_m2m_protocol` validator (intents, pointers)
Swarm 🟡 hardening `forge_swarm` family: NR in CI for 13 of 14 modules (not the base one)
COGNITION
AMI 🟡 active `forge_world_model` + `forge_ami_strategist`
Active Inference 🟡 active `forge_active_inference`; homeostat coupling = prototype
Neuro-symbolic governance ✅ operational `NokidoGateConsumer` + `forge_golden_rules_ast`
Autonomous evolution 🟡 guarded `forge_mutation_judge`: one gate (owner arming, brake, human lock) + auto-brake on capability regression
PHYSIOLOGY
Endocrine ✅ operational `NokidoHormonesListener` + `forge_endocrine`
Nervous system ✅ operational `NokidoAfferent` + `NokidoOrganPulse`
Immune system 🟡 partial `forge_semantic_firewall` + `forge_sovereign_membrane`
Cortex ↔ autonomic loop 🟡 first piece `forge_epistemic_daemon`: epistemic drive (gap → inquiry); full coupling not built
NEURAL SUBSTRATE
Software SNN ✅ experimental `forge_snn_core` + `forge_snn_router`, on demand
NPU / edge 🟡 in development `NokidoBrainWorker`, disabled by default since 2026-07-24
Neuromorphic hardware 🔬 future —
Nokido is an alpha-stage research and engineering project; what it does not claim (AGI, guaranteed self-healing, production stability for every subsystem…) is written in the full README.
🔐 Security
Secrets live in a machine vault (DPAPI · Keychain · libsecret), never in .env. Callers get an integrity ring (six-ring
RBAC); outbound text crosses the SemanticFirewall and the SovereignMembrane before any cloud model; the hub binds
to localhost and cloud access is opt-in. Read SECURITY.md before exposing anything beyond localhost,
and disclose vulnerabilities privately first. Model: 07 — Security ·
08 — Vault and secrets.
📚 Documentation
- Full README — idea, organism, protocols, installation paths, roadmap, principles
- Wiki index (EN / FR) · What's new
- Architecture · Hardware roadmap
MANIFESTO.md(French original) · English translationCONTRIBUTING.md·CODE_OF_CONDUCT.md·docs/CLA.md
License
AGPLv3-or-later · commercial licensing available — see LICENSE and
COMMERCIAL.md.
Metadata
Release files for nokido-agent 0.20.16
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| nokido_agent-0.20.16.tar.gz | 15.3 MB | With provenance | Details for nokido_agent-0.20.16.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| nokido_agent-0.20.16-py3-none-any.whl | Python 3 | none | any | Details for nokido_agent-0.20.16-py3-none-any.whl |
Total release size: 32.4 MB
Release files / nokido_agent-0.20.16.tar.gz
| Download URL | nokido_agent-0.20.16.tar.gz |
|---|---|
| Size | 15.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ef6356f6faab9a380f5c2212be796782f606f7da453d84a15994b7bdfcade81f
|
|
BLAKE2b-256 checksum How to use checksums |
0c470daef9f46afce85153da6734833f7c47500ae631680d50b465e7d0a0e41f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.
Transparency logRelease files / nokido_agent-0.20.16-py3-none-any.whl
| Download URL | nokido_agent-0.20.16-py3-none-any.whl |
|---|---|
| Size | 17.1 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
1d045aff0cd814dcae6be3ac11276fd5e5ee629d5a81d63861a6022fb913ed9e
|
|
BLAKE2b-256 checksum How to use checksums |
f72564f330f2c056e775cde27b4f5548e76bf9fa7df2449a0fbdbfb8c009d79b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.
Transparency logSecurity for nokido-agent 0.20.16
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 2 files in this release have been uploaded with provenance metadata.
Attestation for nokido_agent-0.20.16.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 1, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.