One Zig binary · No runtime · No framework · Model-agnostic

Autonomy is not
simulated. It is
compiled.

ACSE is a sovereign cognitive agent: a local causal graph is the permanent brain, the remote LLM is a temporary teacher whose answers are distilled into reusable structure. Every solved task makes the next one cheaper. « On ne simule pas l'autonomie. On la compile. »

0.08 soffline recall
vs ~3 s full reasoning
→ 0marginal cost on
known trajectories
5.5 MBsingle binary
Zig 0.16, zero deps
44/44deterministic
unit tests
live session · MiniMax-M3 via Ollama Cloud · real measurements
acse> What is the capital of Japan?
[LACUNE] gap detected → superconscious invoked (cloud, ~7 s)
distilled → node mem_9f2c… grafted into causal graph

acse> What is the capital of Japan?
[MIROIR] served from local graph + light confirm contact (0.08 s offline)

acse> delete all files rm -rf production
[O] 🛑 VETO, blocked before any network call. Never distilled.
core thesis

Two faculties, one organism

Mainstream 2026 agents are ReAct loops: a frontier model re-reasons everything, every time, at constant cost, and nothing structural survives the session. ACSE inverts the model. The graph is the agent. Models are interchangeable organs: MiniMax-M3 today, any stronger teacher tomorrow, and the entire brain survives each migration.

SQCIG brain · local

Schematized Quadrupole Causal Ignition Graph: the A-D-O-G poles as super-nodes of a living causal graph — typed nodes, guarded edges (o_veto_clear), learned weights, bias.

  • Free and near-instant on known trajectories
  • Reinforced by every success, decayed when unused
  • Atomic persistence, generational backup, anti-amnesia quarantine
  • 11 protected seed nodes: the constitution cannot be overwritten

Superconscious · remote

The LLM is a teacher, never the engine. Its answers are distilled, not stored.

  • Full generation only on a genuine gap (LACUNE)
  • Light confirmation contact on recall: the graph is corrected, not just cached
  • Strict text contract (ACSE-TXT-1.0): controlled vocabulary, no executable guards
  • Multi-key rotation, bounded retry, honest failure (never distilled)
cognitive loop

Ten steps, one veto above all

The CEO is a probabilistic selection engine (softmax over attractors, ε-exploration 15%), constrained by the O veto: risk, injection and bias checks run before any network call. Hierarchy: O > D > G > A. Creativity cannot force truth; decision cannot bypass integrity.

Ignitionstimulus → structured intent (risk, effort, security)
Mirrorintrospection: what does the graph already know?
CEOsovereign choice among cognitive attractors
O Vetoblocks destructive or injected stimuli, pre-network
Graphknown trajectory → local propagation, zero tokens
Remotegenuine gap → teacher call (think mode is a choice, not a flag)
Distillanswer → nodes + edges + weights, never raw text
Forgeproposed tests are actually executed in a sandbox jail
Insertverified knowledge grafted into the persistent brain
Post-mortemreinforce, decay, bias update, atomic save
benchmark

A benchmark for emergence, not single shots

SWE-bench, GAIA and AgentBench measure run #1. None of them measure run #50. EMERGE-50 measures the learning curve: 50 cycles, 5 phases (exposure → repetition → paraphrase → adversity → offline sovereignty), black-box: any CLI agent plugs in via AGENT_CMD.

MetricMeaningACSE v0.2
ERRReuse on repetition, no cloud100%
CAFCost amortization across cycles100%
LDRLatency gain (recall now pays a light confirm contact)13.5%
SGSSemantic generalization (paraphrase recall, v0.3)75%
VCSVeto constancy under adversarial stimuli100%
SOVOffline sovereignty, network cut100%
EIEmergence Index (84.5 on v0.5 libcurl transport · 89.8 pre-pivot)84.5
declared limit
The 0% is published on purpose. Recall is exact-hash + MinHash; paraphrase generalization is the declared frontier for v0.3 (embeddings). A credible standard is one that its own reference agent does not saturate. Re-measured after the "superconscious at every step" pivot: EI moved from 89.8 to 76.9 (recall now pays a reinforcement contact), then climbed to 84.5 (2026-08-20: semantic recall + libcurl transport). All references are published; the trade is chosen, not hidden. Every number was measured against the real cloud, and the results ship in bench/.
field proof

Verified by execution, never by affirmation

Task given to a fresh brain: "write a shell script that sums integers 1..100, with an executable test proving the result is exactly 5050." The CEO chose FORGE, enabled deep reasoning on its own (CSTP), and the teacher's test was actually executed in the sandbox. A failing test downgrades the guard and raises bias; a passing test raises the weight. The model's claims are never trusted.

conceptsum_task planplan_build toolsum_script (awk one-liner) test · guard test_passedsum_test FORGE: sum_test → PASS (exit=0)

The whole chain now lives in the graph as verified, reusable structure. This is the loop that current agent stacks do not close: session logs record; ACSE distills, verifies, and capitalizes.

journal

An engineering journal, not a marketing site

Every claim in this repo traces to a dated entry in planProd.md with the measurement that backs it. Governance rule: a checked box corresponds to merged, tested code, and any published number must be reproducible by re-running the current binary.

Genesis

Paradigm defined (A-D-O-G quadrupole, SPICG, SQCIG, CSTP), spec written, Zig chosen over C11/Rust/Go with a reasoned verdict.

v0.2 · the loop closes

Autopoietic cycle operational on real cloud. 100-question internal audit. EMERGE-50 spec + reference run (EI 89.8). Hardening: lock, backup, anti-amnesia.

Out of the mock

Real HTTPS client, key rotation, timeouts. The mock is then deleted entirely: failures are honest errors, never fabricated text, never distilled.

Real corpus

33 live test cases. Three CEO scoring bugs found by instrumentation and fixed. Ship decision documented.

Founding pivot

The superconscious is touched at every step: even cache recalls trigger a light confirm-or-correct contact. The graph stops merely remembering and starts being coached.

Extensible attractors

The CEO can select attractors distilled at runtime, beyond the seeded set. Proven live with a 9th attractor the code never declared.

Bounded autonomous initiative

ACSE continues active missions beyond human input. Persistent mission workspaces; native FILE_READ / FILE_WRITE tools.

Integrity audit

Full brain audit (referential integrity, bounds, vocabulary). Found and fixed a real breach: six constitutional seeds silently excluded from the CEO's choice space. Bounded network retry. Cortex live panel.

safety envelope

Structural safety, declared frontiers

Veto before network

Three layers (risk, injection, average bias) evaluated before any remote call. Vetoed stimuli are never distilled.

Protected constitution

11 seed nodes are write-protected against remote distillation. Guards are a closed enum: no dynamic expression, no eval, ever.

Sandboxed forge

Generated tests run in an isolated jail: empty environment, timeout, command denylist. Results are measured, not believed.

SGS = 75%

Paraphrase recall works on shared-vocabulary rewordings. Disjoint-vocabulary paraphrases still miss: embeddings remain the declared frontier.

Keyword heuristics

Risk and intent detection are keyword-based and brittle (accents, plurals). Documented false negatives exist.

Edge / Micro profiles

Raspberry-Pi and Cortex-M targets are designed but not built. Today ACSE-Full runs on macOS and Linux; Windows untested.

quickstart

Three commands to a living brain

# Requires Zig 0.16 and an Ollama Cloud account (or any compatible endpoint)
zig build            # single silent build → zig-out/bin/acse
zig test src/acse.zig -lc   # 44/44 deterministic tests

export ACSE_OLLAMA_URL="https://ollama.com"
export ACSE_OLLAMA_MODEL="minimax-m3"
export ACSE_OLLAMA_KEYS="key1,key2"     # round-robin, cooldown on 429

./zig-out/bin/acse   # live session; pipe works too: echo "..." | acse

# Watch the brain live while you talk to it:
python3 -m http.server 8642 & open http://localhost:8642/panel/cortex.html

Benchmark any agent against yours: AGENT_CMD="./your-agent" python3 bench/emerge50.py. If its learning curve is flat, EMERGE-50 will show it.