Agent Toolchain

Run, review and ship code written by agents

Four small, self-hosted tools that turn autonomous coding agents into a workflow you can trust: plan, implement, review, validate, and ship — with a human in control at the checkpoints that matter.

See the Tools
Agent Toolchain - tools for running and reviewing coding agents
Coming Soon
Self-hosted.
Provider-agnostic.
Isolated by default.
... Human-approved

Times are changing

Since spring 2026, less and less code is written by hand. It's time to invent new ways to produce quality code and quality applications — in cooperation with coding agents, not despite them.

Four tools, one workflow

Each tool covers one stage of working with autonomous agents — from running them, to checking their work, to keeping the pipeline and the apps they build healthy.

AgentFlow

Orchestrator

Submit a feature request against a real repository and AgentFlow runs the whole loop: plan, ask only genuinely blocking questions, wait for your approval, implement in an isolated container, validate, and hand off for review. Claude Code, Codex, and OpenRouter-backed agents sit behind a common interface, so the provider is a detail, not a commitment.

AgentFlow - workspace dashboard

ReviewFlow

Independent reviewer

A second, independent agent — usually from a different provider than the one that wrote the code — reads every diff before a human does. It flags regressions, scope creep, and half-finished work, and can send fixable feedback straight back to the implementing agent before it ever reaches your review queue.

CIFlow

CI & pipelines

Manages the pipelines that validate what agents produce — builds, tests, and deploy checks — and treats agent runs as first-class CI jobs. When a run fails, CIFlow can route the logs to a triage agent instead of paging a human first.

Caravel

Codebase explorer

Points at a Laravel codebase, reads it — routes, models, jobs, Filament resources, conventions, dead code — and turns what it finds into a browsable web app, so a human or an agent can explore and improve the codebase instead of guessing its way through it.

How it fits together

A request goes in, and a working, reviewed, validated change comes out — with two checkpoints reserved for you.

1
Plan
AgentFlow
2
You approve
Human checkpoint
3
Implement
AgentFlow
4
Validate
CIFlow
5
Review
ReviewFlow
6
Fix
AgentFlow
7
You merge
Human checkpoint

Once merged, the change lands in a Laravel codebase — and Caravel is what you open next to see what actually changed: routes, models, and structure, mapped out and ready to explore before the next task begins.

Coming Soon

Three tools to manage LLM agents, and one app they built

AgentFlow, ReviewFlow and CIFlow are built with X-Go and shaped with Gen-code, the same way as the rest of the Blue-Lila suite — small, readable, self-hosted tools instead of another SaaS dashboard.

Try Gen-Code