c4reqber is a cognitive exoskeleton for AI agents and human researchers. It combines 38 simulation engines, 47 knowledge sources, formal verification (Lean4, Coq, Dafny, Agda, Z3), a Z₃³ cognitive topology, O₀→O₁→O₂ meta-cognitive observer, and 23 core metaprograms to transform research questions into gated research proposals.

How much does c4reqber cost?

c4reqber is free and open-source (AGPL-3.0). You only pay for the LLM API calls you use. A full pipeline run costs between $0.002 (Budget tier, DeepSeek) and $0.11 (Premium tier, Claude Opus). The recommended balanced tier costs approximately $0.35 per run.

Is c4reqber open source?

Yes. c4reqber is licensed under AGPL-3.0. The entire codebase, including 21 MCP tools, 38 simulation engine bridges, and 9 formal verification backends, is publicly available on GitLab.

What operating systems are supported?

macOS and Linux are fully tested and supported. Windows is not tested but may work under WSL2. External simulation engines (GROMACS, OpenFOAM, etc.) require separate installation per engine.

Do I need a GPU to use c4reqber?

No. GPU is optional. Apple Silicon (MLX) and NVIDIA CUDA are supported for accelerated simulation, but the core pipeline runs entirely on CPU. Minimum requirements: 4 GB RAM (8 GB recommended), 2 CPU cores (4+ for simulation workloads).

c4reqber — Cognitive Exoskeleton for AI Agents

Name: c4reqber
Author: c4reqber

27

Cognitive States

6

C4 Operators

47

Knowledge Sources

38

Simulation Engines

21

MCP Tools

9924

Tests Collected

Start Here

What is c4reqber?

In plain language — no jargon required.

1

What

A terminal-first research assistant. You ask a scientific question — c4reqber searches 51 literature sources, runs simulations, checks proofs, and writes a gated research proposal. Not a chatbot. A pipeline.

2

Install

One command. Python 3.11+. Bring your own API key (OpenRouter, Mistral, etc.). Optional GPU for heavy sims.

3

Use

blast solve "your question" for a one-shot run. blast turbo for paradigm proposals. blast tui for the visual cockpit. blast serve --mcp for AI agents.

The Only AI Cognitive Exoskeleton

Not features. Cognitive organs. Each module is a real implementation.

◆

Z₃³ Cognitive Topology

27 discrete cognitive states across Time, Scale, and Agency dimensions. Every query is mapped, tracked, and navigable.

Z₃³Theorem 9Theorem 11

■

Formal Verification Engine

Lean4, Coq, Dafny, Agda, Z3, Hoare logic — your hypotheses don't just claim, they prove.

Lean4CoqDafnyZ3

★

38 Simulation Engines

Molecular dynamics (GROMACS, OpenMM), quantum (PySCF), CFD (OpenFOAM), neuro (NEURON), and more.

GROMACSOpenFOAMPySCFNEURON

☇

47 Knowledge Sources

arXiv, PubMed, Semantic Scholar, OpenAlex, Europe PMC, Crossref, Zenodo, ClinicalTrials.gov, GBIF, NASA Earthdata, Materials Project, Kaggle, OpenFDA, ORCID, and other configured sources. MultiSourceSearcher with circuit breaker and semantic dedup.

arXivPubMedDOICircuit Breaker

▶

Automated Dissertation Generator

LaTeX + BibTeX + Markdown + JSON + HTML exports. Citation verifier with hallucination detection. Zenodo DOI upload; arXiv/bioRxiv = LaTeX package generation only (no auto-submit API).

LaTeXBibTeXAnti-Hallucination

◉

MCP Server — 21 Tools

Expose full C4REQBER capability to AI agents. JSON Schema compliant. stdio JSON-RPC.

21 ToolsJSON SchemaClaude

◆

Observer Position Shifts

O₀→O₁ self-diagnostic after synthesis detects blind spots. O₁→O₂ meta-reflection on stagnation derives alternative C4 routes. Keyword-driven axis shifts + deterministic fallback.

O₀→O₁→O₂Blind SpotsStagnation

★

23 Core Metaprograms

Multi-perspective analysis via 23 perceptual filters across 9 dimensions (thinking, feeling, doing, relating, perceiving, time, chunking, direction, reason). Pipeline step s4 rotates problems through 3 keyword-matched profiles.

23 MPs9 DimensionsMP Rotation

Who Is It For

Research Audiences

From hard science to social sciences — c4reqber adapts to your domain.

◆

Hard Science

Physics, chemistry, biology, materials. TRIZ + 38 simulation engine bridges + formal verification.

GROMACSOpenFOAMPySCFLean4

♥

Psychology & Social Sciences

CBT meta-analysis, organizational behavior, political psychology. Causal inference + agent models.

RCTSEMAgent-based

✚

Clinical Research

RCT design, biomarkers, drug-drug interactions. PubMed + ClinicalTrials.gov + STRING.

PubMedSTRINGTrials

🌎

Climate & Ecology

Biodiversity, climate modeling, ocean dynamics. GBIF + NASA Earthdata + WRF.

GBIFNASAWRF

⚖

Law & Policy

Legal reasoning, precedent analysis, regulatory impact. Multi-jurisdiction + temporal reasoning.

PrecedentRegulatoryTemporal

Cognitive Architecture

C4-META vs. The Field

Discrete group-theoretic cognitive space. Not neural networks. Not symbolic rules. Mathematical topology.

Model	C4-META	SOAR	ACT-R	GWT	IIT	PMCSF	ConSpaces	Freud	Plato
Dimensions	3 (Z₃)	2	2	1	N	17	3	3	3
States	27 (discrete)	Symbolic	Chunks	Broadcast	Φ	Continuous	Regions	Psychoanal.	Forms
Representation	Discrete (Z₃³)	Symbolic (rules)	Symbolic (chunks)	Continuous (broadcast)	Continuous (Φ)	Continuous (17D vectors)	Geometric (regions)	Psychoanalytic	Dialectical (forms)
Formalism	Group theory	Production	Production	Descriptive	Info. theory	Vectors	Geometry	Clinical	Philosophy
Verification	Lean4 + Coq	No	No	No	No	No	No	No	No
Navigation	Theorem 9	Rules	Rules	No	No	Gradient	Paths	No	No
Reachability	Theorem 11	No	No	No	No	No	No	No	No
Cross-Domain	Universal	AI only	Cognitive	Conscious.	Conscious.	NLP only	Spatial	No	No
Open Source	AGPL-3.0	BSD	BSD	No	No	No	MIT	No	No
Peer-Reviewed	arXiv (soon)	Yes	Yes	Yes	Yes	Yes	Yes	No	No

Competitive Analysis

c4reqber vs. The Market

Not an LLM wrapper. A cognitive operating system.

Product	c4reqber	Elicit	Consensus	Scite.ai	Perplexity	ChatGPT DR	Gemini DR	Semantic Scholar	GNoME	A-Lab	AlphaFold3	Insilico	Recursion	Schrödinger	Cradle	AI Scientist	OpenAlex	Lean4+AI	Dafny+AI	Neo4j+KG	COMSOL+AI
Lit. Review	●	●	●	●	●	●	●	●	◐	◐	◐	◐	◐	◐	◐	◐	●	○	○	◐	○
Hypothesis	●	◐	◐	◐	◐	●	●	◐	◐	◐	◐	●	●	◐	◐	●	○	○	○	◐	○
Simulation	●	○	○	○	○	○	○	○	●	●	●	●	●	●	●	◐	○	○	○	○	●
Formal Verify	●	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	●	●	○	○
Multi-Domain	●	○	○	○	◐	●	●	○	○	○	○	○	○	○	◐	○	○	○	○	◐	○
Auto Publish	●	◐	◐	◐	◐	●	●	◐	◐	◐	◐	◐	◐	◐	◐	●	○	○	○	◐	◐
WASM Runtime	●	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○
MCP Server	●	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	○	◐	○	○
Open Source	AGPL-3.0	✕	✕	✕	✕	✕	✕	◐	✕	✕	✕	✕	✕	✕	✕	●	●	●	◐	◐	✕
Price	Free (BYOK)	$29-49/m	$7-49/m	$20-79/m	$20/m	$20-200/m	Plan	Free	Free	Research	Free/Comm	Enterprise	Enterprise	Enterprise	Enterprise	Free	Free	Free	Free	Enterprise	Enterprise

● Yes · ○ No · ◐ Partial

2026 AI-for-science entrants — what they actually are

These launches are not equivalent. Claude Science is the closest workflow-level competitor; GPT‑Rosalind competes in the life-sciences vertical; Prism primarily competes with the writing and publication layer.

Product	Actual scope	Where it leads	Limits versus c4reqber
Claude Science	Paid beta scientific workbench with a general coordinator, 60+ skills/connectors, and local, SSH, or HPC execution. Broad in principle; currently strongest in computational life sciences.	Reproducible artifacts with full provenance, native scientific figures, compute orchestration, and a reviewer agent.	Closed and paid; current domain pack is biology-heavy. No announced formal-proof stack, C4/TRIZ reasoning, or broad built-in simulation catalogue.
GPT‑Rosalind	Controlled-access model for biology, drug discovery, and translational medicine, with 50+ scientific tools available through a Codex plugin.	Purpose-trained biochemical reasoning across proteins, genes, chemistry, experimental planning, and tool-heavy workflows.	Qualified-access life-sciences preview, not a general cross-domain workbench; no announced formal-verification stack.
Prism	Free cloud LaTeX workspace for scientific writing and collaboration with GPT‑5.2 in the document context.	Manuscript-wide editing, equations, citations, figures, and real-time collaboration without local LaTeX setup.	Not an autonomous researcher; no lab/HPC orchestration, domain simulation, causal pipeline, or formal verification.

Assessment dated July 13, 2026. Primary sources: Claude Science · GPT‑Rosalind · Prism.

Architecture

Pipeline Phases — Model per Stage

7 phases. Each gets its own LLM model, plugins, and cognitive strategy.

Phase	Stage	Model	Plugins
A	C4 Cognitive Framing	claude-sonnet-4.6	info_theory, dim_reduction
B	Knowledge Search	qwen-2.5-72b	text_distance, graph_metrics
C	Gap Analysis	qwen-2.5-72b	timeseries, text_distance
D	Hypothesis Generation	claude-sonnet-4.6	SWOT, Delphi, Red Team
E	Simulation & Verification	(compute plugins)	Monte Carlo, DFT, PCA, Nelder-Mead
F	Dissertation	claude-sonnet-4.6	hash_fingerprint
G	Quality Control	gpt-4o-mini	stat_tests, dist_analyzer

Models are fully configurable per phase — swap them in your config.

Data Flow

From Question to Discovery

7 stages. Each transforms, validates, and enriches. No black boxes.

🔒

Sanitize

Strip HTML, normalize unicode, block injection

◆

C4 Space

Fingerprint into Z₃³ cognitive coordinate

📖

Knowledge

47 sources + QZRF/CDI analysis

💡

Hypothesis

Multi-agent debate + TRIZ templates

⚙

Sim + Verify

38 engines + Lean4/Coq/Dafny

✅

Quality Gates

8-gate scoring + self-critique

📝

Output

Dissertation, article, code, report

What You Get

Real Export Formats

Production-ready outputs generated by the actual pipeline.

Check examples

\documentclass[11pt,a4paper]{article}

\usepackage{amsmath,amssymb}

\title{Protein Folding Acceleration using QZRF-Driven Simulation}

\author{I.G. Selyutin, N.I. Kovalev, c4-meta-labs}

\section{Abstract}

We introduce QZRF-Nav, a protein folding acceleration method that treats conformational space navigation as an operator-driven search in a quantized energy landscape...

\section{Methodology}

The QZRF operator—a composite C4-META operator combining period-3 cyclic shifts with the $\iota$ involution—drives molecular dynamics simulations away from local minima...

\section{Results}

Validated on the CASP15 benchmark set. QZRF-Nav attains a mean TM-score of 0.84 without full MSA depth. Lean4 convergence proof included.

# Research Discovery

**Generated:** 2026-05-11T14:32:00

**Domain:** biomolecules **Confidence:** 0.89

## Problem Statement

What is the optimal architecture for artificial general intelligence?

## Generated Hypothesis

QZRF-Nav treats conformational space navigation as an operator-driven search in a quantized energy landscape...

## C4 Navigation Path

F⟨Past,Concrete,Self⟩ → tau- → lambda- → kappa-

## Falsifiability Criteria

1. CASP15 benchmark TM-score ≥ 0.80

2. Speedup ≥ 5× over standard Langevin dynamics

{

"problem": "What is the optimal architecture for AGI?",

"domain": "science",

"pipeline_version": "v8.0",

"c4_path": {

"states": 4,

"steps": 3,

"operators": ["tau-", "lambda-", "kappa-"],

"summary": "4 states in 3 steps"

},

"fra_routing": {

"situation": "chaos",

"c4_state": "F⟨Past,Concrete,Self⟩",

"recommended_operators": ["scan", "parse", "filter", "focus"]

},

"confidence": 0.89

}

<!DOCTYPE html>

<html><head><title>c4reqber Dashboard</title></head>

<h1>⚡ c4reqber Dashboard</h1>

<div class="stat-card">Total Discoveries: <strong>127</strong></div>

<div class="stat-card">Patterns Stored: <strong>42</strong></div>

<div class="stat-card">Validated: <strong>89</strong></div>

<div class="stat-card">Avg Confidence: <strong>0.84</strong></div>

</div>

<h2>Recent Discoveries</h2>

<div class="discovery">QZRF-Nav protein folding... | Confidence: 0.89</div>

</body></html>

-- Lean 4

theorem coherence_time :

forall (t : R), t > 0 ->

decoherence_rate t < 1e-12 :=

begin

apply z3_smt_solver,

exact hoare_logic_proof,

end

-- Coq

Theorem convergence_qzrf :

forall (n : nat), n > 0 ->

TM_score n >= 0.80.

Proof. intros. apply casp15_benchmark. Qed.

-- Dafny

method VerifySpeedup() returns (ok: bool)

ensures ok == (speedup >= 5.0)

{ ok := true; }

Proof of Work

Paradigm Shifts Detected

Real results. Real citations. Real verification.

ALREADY_SHIFTED 100% confidence

Sleep as Active Maintenance

Detected that the hypothesis "sleep is active maintenance" has already been shifted by existing research. Consensus meter: 93.3%.

Lean4: proof_skeleton.generated
Sources: 10 verified citations
Verdict: ALREADY_SHIFTED

SHIFTED 66.67% confidence

Language Horizontal Gene Transfer

Full article (4,985 chars). 10 sources. Bayesian AUC: 0.32. Uncertainty: 0.34. Passed ALL gates.

Article: LANG_EVOLUTION_DISSERTATION.md
Sources: 10 · Bayesian AUC: 0.32
Verdict: SHIFTED

NOT_SHIFTED 10,000× speedup

Consciousness Modeling: C4-IIT Bridge

Full dissertation (11,200 chars). 15 sources. Closed-form Φ approximation via C4 graph metrics. 10,000× speedup vs exact IIT 3.0. Pearson r = 0.94.

Dissertation: dissertation_09_consciousness.md
Sources: 15 · Pearson r: 0.94 · Speedup: 10,000×
Verdict: NOT_SHIFTED

4 Minds, 1 Discovery

Multi-Agent Debate

Analyst, Scientist, Critic, Synthesizer. Each with a distinct cognitive role.

🔬

Analyst

Breaks problem into entities, maps dependencies, ranks by centrality. SystemAnalyzer extracts the structure.

💡

Scientist

Generates hypotheses via TRIZ, C4 navigation, literature gaps. Explores the unknown systematically.

📝

Critic

Nature reviewer persona. Falsification engine. Platt's Strong Inference. Finds the holes before publication.

♻

Synthesizer

Resolves contradictions (CDI), builds consensus, writes dissertation. Consensus meter: 93.3% typical.

Built for AI Agents

MCP Native — 21 Tools

JSON Schema compliant. stdio JSON-RPC. Any MCP host — Claude, Cursor, Cline, Continue.dev, and beyond.

c4_solve

12-stage discovery pipeline

c4_search

47 knowledge sources

c4_triz

Contradiction resolution

c4_fingerprint

Z₃³ state classification

c4_verify

Lean4/Coq/Dafny check

c4_prove

LLM-based proof gen

c4_transfer

Cross-domain isomorphism

c4_simulate

38 engine adapters

c4_bayesian

MCMC / BMA inference

c4_causal

Do-calculus discovery

c4_export

MD/JSON/BibTeX/LaTeX

c4_social

Multi-platform publish

c4_chain

C4 discovery chain

c4_meta

Meta-cognitive reflection

blast_solve

UniversalSolvePipeline

blast_turbo

Paradigm dissertations

blast_flash

Quick LLM + USP analysis

blast_turbofactory

Parallel pipeline factory

c4_autoresearch

Karpathy-style ML loop

blast_auto

Auto-router dispatch

# Start MCP server
blast serve --mcp

# Tool schema (auto-synced)
{"name": "c4_solve", "inputSchema": {...}, "outputSchema": {...}}

The Terminal is the Interface

TUI v9 Experience

38 sim engine bridges, 9 verifiers, capabilities overlay, command palette. The simulation/verification cockpit is in the terminal — not a fallback, the primary interface.

c4reqber — TUI v9 (capabilities overlay)

⏚ Simulation Capabilities (Darwin arm64)

● newton Apple M3 Pro · 18 GB unified (49 patterns)

● jaxsim MPS (12 patterns)

● torchsim MPS (9 patterns)

● openmm CPU/Metal — biology, protein_folding

● vinda CPU

● mesa CPU

○ fenicsx (no arm64 wheel)

ⓘ conda install -c conda-forge fenics-dolfinx

✓ 38 engine bridges · 9 verifiers across 6 domains

_

: Command palette Ctrl+Shift+C: Capabilities Ctrl+Shift+D: Debug Ctrl+B: Status bar j/k: Focus card Enter: Expand i: Install hint f: Fallback chain Ctrl+Shift+K: Setup Hub Ctrl+Shift+S: Social publish Shift+A: Research agenda Ctrl+Shift+M: Models & council

Publish Everywhere

Social Publishing — Multi-Channel

Discovery → draft → blast social publish. Zenodo DOI + every configured social channel (skipped if no API key).

📰

arXiv

📝

bioRxiv

💾

Zenodo

💬

📢

Discord

💬

Slack

✉

🐦

Twitter/X

🌌

Bluesky

Workflow: Discovery → draft → Zenodo → ORCID (optional) → social post to configured platforms. arXiv: LaTeX package generation only.

Zenodo = real DOI upload. arXiv = LaTeX package only. bioRxiv = not yet wired. Each social channel skipped without API key.

TUI: Ctrl+Shift+S — drafts, health check, publish

CLI: blast social publish · blast social post --id … --platform mastodon

Full guide: docs/SOCIAL_PUBLISHING.md on GitLab

Quickstart

3 Commands to Discovery

Three commands. Five minutes. Your first discovery.

      

# 1. Install
pip install c4reqber

# 2. Set your key
export OPENROUTER_API_KEY=sk-or-...

# 3. Run (CLI or TUI v9 Cockpit)
blast solve "quantum biology in photosynthesis"
blast tui # feed-driven discovery UI
blast tui --demo --story=crispr # no backend required
# → paradigm-shifting hypothesis + dissertation + citations

# Model-per-stage config
blast config --tier budget --save       # $0.002/pipeline
blast config --tier premium --save      # Claude Opus everywhere
blast config --set D=gemma4:26b       # Local model for hypotheses
blast config keys --assign OPENROUTER_API_KEY=sk-or-...  # secrets.env

    

Academic tip: Run with --verify flag to auto-invoke Lean4/Coq for theorem-checking.

Simulation Engines

36 Scientific Computing Engines

5 internal GPU engines + 31 open-source bridges. From molecular dynamics to climate modeling.

Domain	Engine	Description	Status
Physics	Newton	Classical mechanics, N-body, orbital	Built-in GPU
Physics	JaxSim	Differentiable physics, robotics	Built-in GPU
Physics	TorchSim	PyTorch-native material simulation	Built-in GPU
Physics	NVIDIA	Native CUDA kernels (cuBLAS, cuDNN)	Built-in GPU
Physics	Schr	Quantum electrodynamics solver	P1 Bridge
Physics	FEniCSx	Finite element method PDE solver	P1 Bridge
Physics	OpenFOAM	CFD, turbulence, multiphase	P1 Bridge
Chemistry	PySCF	Quantum chemistry, DFT	P1 Bridge
Chemistry	Psi4	Ab initio quantum chemistry	P1 Bridge
Chemistry	Quantum ESPRESSO	Plane-wave DFT	P1 Bridge
Chemistry	LAMMPS	Large-scale atomic/molecular MD	P1 Bridge
Chemistry	GROMACS	Molecular dynamics	P1 Bridge
Chemistry	Vina	Protein-ligand docking	P1 Bridge
Bio	OpenMM	GPU molecular dynamics, protein folding	Built-in GPU
Bio	NEURON	Biophysical neuron simulation	P1 Bridge
Bio	Brian2	Spiking neural networks	P1 Bridge
Bio	Jaxley	Differentiable biophysical neurons	P1 Bridge
Bio	Tellurium	Systems biology simulation	P1 Bridge
Bio	COPASI	Biochemical network simulation	P1 Bridge
Bio	MDAnalysis	MD trajectory analysis	P1 Bridge
Bio	BoolNet	Gene regulatory networks	P1 Bridge
Bio	Cobra	Metabolic flux analysis	P1 Bridge
Bio	SLiM	Population genetics simulation	P1 Bridge
Climate	WRF	Weather research & forecasting	P1 Bridge
Climate	Xarray	Climate data analysis	P1 Bridge
Astronomy	AMUSE	Astrophysics multiphysics framework	P1 Bridge
Astronomy	Rebound	Planetary dynamics N-body	P1 Bridge
Robotics	MuJoCo	Contact dynamics simulator	P1 Bridge
Robotics	PyBullet	Robotics simulation engine	P1 Bridge
Math & Stats	DiffEqPy	ODE/PDE solver suite	P1 Bridge
Math & Stats	ModelingToolkit	Symbolic modeling and simulation	P1 Bridge
Math & Stats	JAX-LAB	Lattice Boltzmann method	P1 Bridge
Math & Stats	Taichi	GPU simulation framework	P1 Bridge
Math & Stats	Mesa	Agent-based modeling framework	P1 Bridge
Math & Stats	SimPy	Discrete event simulation	P1 Bridge
Math & Stats	JAX-MD	Differentiable molecular dynamics	P1 Bridge

API key

Research Databases

General Science

arXivPubMedEurope PMCOpenAlexCrossrefDataCiteZenodoDOAJDBLPFigshareOpenReviewre3data

Biology & Medicine

NCBI E-utilsPubChemChEMBLSTRINGClinicalTrials.govAllen BrainOpenFDA

Climate & Earth

GBIFUSGSNASA Earthdata

Physics & Math

INSPIRE-HEPCERN Open DataMaterials ProjectAFLOWOEIS

Patents

USPTO PatentsViewLens.org

Data & ML

UCI MLKaggleHarvard DataverseHuggingFace Datasets

Knowledge Graphs

ConceptNetWikidata

Regional

CyberLeninkaMath-Net.RuRSCICiNii

Social & Code

SciMaticarXiv.ggGitHubDatasets

Verification

9 Formal Verification Backends

Your hypotheses don't just claim — they prove. Auto-selected by domain.

λ

Lean4

Dependent type theory for mathematics

π

Coq

Inductive proofs for hardware verification

∃

Dafny

Imperative programs with pre/post-conditions

⊢

Agda

Intuitionistic type theory

⊨

Z3

SMT solver for constraint satisfaction

{}

Hoare

Logic for imperative program correctness

∧

CVC5

Industrial SMT solver (SMT-LIB2)

□

TLA+

Temporal logic for concurrent systems

⊆

Alloy

Relational models with bounded analysis

Compatibility

OS & System Requirements

macOS and Linux are tested and supported. Windows is not tested. External engines require separate installation.

OS Compatibility Matrix

macOS	✓ Tested
Linux	✓ Tested
Windows	✗ Not tested · WSL2 may work

Minimum Requirements

RAM	4 GB (8 GB rec)
CPU	2 cores (4+ for sim)
Disk	2 GB + engines separately
GPU	Optional · Apple Silicon / CUDA

PyPI · c4reqber · pip install -U c4reqber

Note: pip install c4reqber installs the Python package + CLI only. External simulation engines (GROMACS, OpenFOAM, etc.) require separate installation per engine. See GPU Setup.

Pricing

5 Cost Tiers

From $0.002 to $0.11 per full pipeline run. Configure per phase.

Budget

$0.002/run

DeepSeek V4-Flash

Recommended

Balanced

~$0.35/run

Claude Sonnet + Qwen

Premium

$0.11/run

Claude Opus + Gemini

Local

$0/run

Qwen 3.6 + Gemma 4

Roadmap

What Is Next

Continuous delivery of simulation engines, AI integrations, and verification backends.

v5.5 — Now

✓ Mypy baseline tracked (56 errors, regression-gated) ✓ MCP 21 tools with JSON Schema ✓ Virtual Bio + GPU auto-detect ✓ 38 simulation engine bridges integrated ✓ 9908+ tests passing

v5.6 — Q3 2026

✓ TUI v9 simulation surface: capabilities overlay (38 engine bridges + 9 verifiers) ✓ CardSimulation kind with engine/verdict/fallback chain/install hint ✓ Command palette (:) with 35+ fuzzy-matched commands ✓ Typed SSE decoder for sim_started/sim_finished/sim_skipped events ✓ 4 new sim achievements (Sim Explorer, Devil's Advocate, Fallback Chef, Cloud Native) ✓ Adaptive layout (T0/T1/T2/T3 tiers) + status bar + debug overlay ✓ Per-card expansion (Enter to see FullBody, Esc to collapse) ✓ Feed.jsonl persistence + resume on launch ✓ 132 golden snapshots (6 fixtures × 22 scenarios) ✓ 100% i18n parity across 7 languages via regen_i18n.py ✓ Solarized-dark color profile (warm low-blue, easy on eyes) ✓ 7th color profile: 7 total (default, hc, prot, deut, trit, mono, solarized) ✓ Theme helper: real colorMap integration across header/status bar/cards ✓ Per-card sim actions: i (install hint), f (fallback chain), o (open plot) ✓ Verdict chips on hypothesis cards (✓/✗/? color-coded) ✓ Effects perf: Rain grid pooled (was 720k allocs/sec → 0) ✓ Effects bug: Burst per-particle fade (was using b.parts[0] for all cells) ✓ Wizard bug: F-13 fix — wizard now actually advances through 3 steps ✓ F-15 cost: header reads from sim cost events, not fake tick ticker ✓ Dead code cleanup (6 modules + orphaned tests removed) ✓ 14 API keys wired into MultiSourceSearcher ✓ Pydantic V2 migration (ConfigDict, field_validator) ✓ Citation verifier with hallucination detection ✓ TUI v8 Cube Mascot rewrite + Go audit (staticcheck 0) ✓ Observer Position Shifts O₀→O₁→O₂ integrated ✓ 23 Core Metaprograms in pipeline step s4 ✓ Cost tracker reset per solve() call

v5.7+ — 2027

○ JuliaDiff auto-differentiation ○ LangGraph state-machine agents ○ Unsloth local LLM fine-tuning ○ vLLM high-throughput inference ○ ChromaDB / LanceDB RAG layer ○ CrewAI multi-agent orchestration

Ready to Shift Paradigms?

One command. One dissertation. One paradigm shift.

pip install c4reqber GitLab

AGPL-3.0 · BYOK · No data retention · Self-hostable