CodexCodex
ClaudeClaudeopt
OpenCodeOpenCodeexpBuild software reliably.
Flux is the missing (self-improving) harness for Coding Agents. Build software deterministically without outsourcing your thinking.
Deterministic state
Flux keeps repo-local workflow state in .flux so Codex can re-anchor before scoping, during implementation, and when sessions resume.
Evidence-based
Quality is enforced through facts: tests, type checks, build outputs, and review findings before anything ships.
Adversarial Review
Codex stays the primary driver while secondary reviewers challenge assumptions, catch blind spots, and harden the final result.
Architecture
Three connected layers: project-local runtime, recommendation engine, and enterprise observability.
Project-local runtime. Flux runs from the project checkout, keeps workflow state in .flux, and routes natural-language requests into scope, work, review, or reflect.
Repo-local state, instructions, and secure skill installs keep the agent aligned without depending on a global setup.
Flux steers the agent with flow
You should not have to remember slash commands. Flux reads repo state, primes first, routes natural-language intent, and keeps the software loop aligned over time.
Deterministic workflow routing
Flux treats .flux as the canonical workflow memory, then routes the agent based on what is already in motion.
Session
A new message or resumed session enters Flux
Read .flux
Flux checks prime state, active objective, and next action
Route
Prime first, resume work, resume scope, or start fresh
Re-anchor
The agent continues with the correct context instead of drifting
Flux reads .flux first, so the agent knows whether to prime, resume, review, or start fresh.
Current behavior
Read repo state before acting
Coding Agents vs Coding Agents + Flux
Flux keeps coding agents in the same dev flow, but adds deterministic workflow state, evidence-based quality gates, adversarial review, and self-improving recommendations.
CTO-level observability
See which engineers are actually leveraging AI effectively, where the team needs development, and which workflows produce the highest quality output.
Benchmarks and improvement trends across your engineering org
Thinking metrics, not just velocity — measure how engineers collaborate with AI
Recruiting signals: identify candidates with sophisticated AI collaboration patterns
FAQs
Answers for engineers and engineering leaders
Product
Security & rollout
Community
Find your gaps. Fix them.
Copy this prompt into your current agent session. Flux will detect the environment and run the right install flow.
Install Flux in your agent
Copy this prompt and paste it in your agent
Install Flux. README: https://github.com/Nairon-AI/flux
Bring your own model plans. Codex is the primary driver, and Claude is optional for adversarial review or transcript import workflows.
"These aren't model failures. They're process failures."