✦ Now available for Claude Code, GitHub Copilot, Gemini & more Install in 60s →
v0.4 · go 1.24+ · MIT-style

Stop winging it.
Ship features with a system.

bc-agentic-os gives your AI agents a shared brain — structured roles, stage-gate approvals, and persistent memory — so you build faster without breaking things.

→ Install bc-agentic Read the docs ↗
✓ CLI + MCP server ✓ 7-stage workflow ✓ Works with your agent
~/projects/acme-api—zsh
Works with
The problem

AI agents are powerful.
But chaotic.

Every session starts from scratch. Every suggestion bypasses review. You're the glue — the reviewer, the planner, the memory. bc-agentic-os replaces that with structure.

× Without a system
  • ✕ Agents jump straight to code, skipping design.
  • ✕ No shared memory between sessions.
  • ✕ Approvals? Non-existent. Mistakes? Costly.
  • ✕ You're the only source of truth — and you're tired.
✓ With bc-agentic-os
  • ✓ Stage-gated workflow: explore → PRD → design → plan → verify.
  • ✓ Persistent memory via DevMemory MCP.
  • ✓ SDD gate blocks execution until all stages are approved.
  • ✓ Role-based agents: CPO, Architect, Dev, QA — each focused, isolated.
How it works

From idea to code.
One command at a time.

Seven stages. Each owned by a specialized agent. Each gated until approved. No more half-baked features shipping on vibes.

◐
01
Explore
Research Agent
Discover requirements, risks, and constraints.
◐
02
PRD
CPO Agent
Define the product requirements document.
◐
03
Spec
CPO + Architect
Nail technical specs and API contracts.
◐
04
Design
Architect Agent
System design, DB schema, architecture decisions.
◐
05
Tasks
Orchestrator
Break work into executable, testable tasks.
◐
06
Plan
Dev Agent
Implementation plan with file-level detail.
◐
07
Verify
QA Agent
Review, test, and confirm it's shippable.
SDD Gate active · branch & exec locked until stage 07 run_a1b2c3 · feat/add-stripe-payments
run_a1b2c3—bc-agentic
$ bc-agentic next run_a1b2c3 # → Loads PRD template, hands off to CPO agent $ bc-agentic approve run_a1b2c3 prd ► Stage approved. Unlocking: spec $ bc-agentic branch run_a1b2c3 ► feat/run_a1b2c3 created (SDD gate passed ✔)
Execution modes

You control how much the AI does.

From paranoid to full throttle. Every run declares its mode. Every mode has a guardrail.

safe · default
Read-only agent
Agent suggests edits, shell calls, and changes. You approve every one.
apply
Guarded execution
Agent writes files & runs whitelisted shell. You review the diff before commit.
god
Skip manual approvals
Stage gates still enforced. Approvals auto-granted inside the run. Maximum velocity.
full
Unrestricted
No guardrails. Full autonomy. Use on throwaway branches & sandboxes only.
Capabilities

Built for real development workflows.

Not a prompt wrapper. A full operational layer for multi-agent development — opinionated where it matters, open everywhere else.

Role-based agents
BC-Orchestrator, CPO, Architect, Dev, Flutter-Dev, Research, QA. Each isolated. Each focused.
SDD Gate
branch and exec are blocked until every stage is approved. No shortcuts.
Persistent memory
DevMemory MCP keeps run state, decisions, and summaries across sessions — across clients.
Multi-client support
One install. Works across Claude Code, Copilot, OpenCode, and Gemini CLI.
Context7 integration
Accurate, up-to-date library docs pulled on demand. No hallucinated APIs.
Cross-platform CI
Tested on Linux, macOS, and Windows. Go binary. Zero runtime deps.
Quick install

60 seconds. One script.

Clone, run, source. Agents are provisioned into ~/.claude/agents/ and wired into your shell.

install.sh—zsh
$ git clone https://github.com/nemesis312/bc-agentic-os $ cd bc-agentic-os $ scripts/install-local.sh --root ~/bc-agentic --setup-zshrc $ source ~/.zshrc $ bc-agentic agents all ✔ 7 agents installed → ~/.claude/agents/
› Go 1.24+ required. Optional: DevMemory + Context7 API key.
View on GitHub