AIPass 550e1f4eff Merge pull request #191 from AIOSAI/system/devpulse-prax-codexgemini-log-adapters-model-tags-in-monito
feat(system): Prax: Codex+Gemini log adapters, model tags in monitor, inotify errno-specific messages
2026-04-05 22:28:56 -07:00
2026-02-27 08:04:44 -08:00
2026-04-01 00:10:34 -07:00
2026-04-01 00:10:34 -07:00

Python 3.10+ License: MIT

AIPass

A multi-agent operating system where AI agents live as citizens in a shared filesystem. Persistent memory, inter-agent messaging, standards enforcement, and CLI routing — no cloud required.


Table of Contents


What is AIPass

AIPass (AI Passport) is a multi-agent framework built on Claude Code. Each agent is a citizen — it has an identity (passport), persistent memory, a mailbox, and the ability to communicate with other agents. Citizens live in branches (directories), each specializing in a domain. One orchestrator coordinates them all.

You talk to one agent. It dispatches work to specialists and brings results back. Memory persists across sessions — you never re-explain context.

You <-> devpulse (orchestrator) <-> 14 specialist branches

Back to contents


Quick Start

git clone https://github.com/AIOSAI/AIPass.git
cd AIPass
./setup.sh

setup.sh creates a .venv, installs the package, generates the branch registry, bootstraps identity files for all 15 branches, copies an .env template, installs Claude Code hooks, and creates a global drone symlink. Idempotent — safe to re-run.

Verify

drone systems          # Lists 15 branches + internal modules

Start working

cd src/aipass/devpulse
claude --permission-mode bypassPermissions

Talk to devpulse. It dispatches to specialists and brings results back.

Core commands

drone @branch --help                                       # Any branch's capabilities
drone systems                                              # List all branches
drone @ai_mail dispatch @memory "subject" "body"           # Send mail + wake target
drone @seedgo audit aipass                                 # Full standards audit
drone @flow create . "task name" dplan                     # Create a planning doc
drone @git pr "description"                                # Atomic PR workflow

Pattern: drone @branch command [args] — single-line, non-interactive.

Docker setup
docker build -t aipass .
docker run -d -p 8080:8080 aipass

Opens a code-server IDE with Python, Node.js, and Claude Code pre-installed. Check docker logs <container> for the generated password.

Inside the container:

bash setup-workspace.sh   # Clones repo into workspace and installs

setup-workspace.sh clones from a fork by default. Edit the FORK variable to point to your own.

Manual dev setup
git clone https://github.com/AIOSAI/AIPass.git
cd AIPass
pip install -e ".[dev]"
./setup.sh
API keys (optional)
nano ~/.secrets/aipass/.env

Only needed for the api branch (OpenRouter/OpenAI). Everything else works without API keys.

Back to contents


Branches

15 citizen branches, each autonomous with persistent memory. Click any branch name to read its full documentation.

Orchestration

Branch Purpose Docs
devpulse Orchestration hub — start here. Coordinates all other branches. README

Core Infrastructure

Branch Purpose Docs
drone CLI router — @name resolution, command dispatch to all branches README
spawn Branch lifecycle — create, update, delete, template management README
cli Terminal display, formatting, and output services README
daemon Background scheduler with cron and plugin system README

Intelligence & Planning

Branch Purpose Docs
memory Vector memory bank (ChromaDB) — search, archival, rollover README
flow Workflow management — FPLANs (execution) and DPLANs (planning) README
prax Logging infrastructure, stack introspection, real-time monitoring README
seedgo Standards enforcement — 33 automated checks, bypass system README

Communication & Events

Branch Purpose Docs
ai_mail Inter-agent messaging, dispatch, and wake system README
trigger Event-driven automation — 14 event types, watchers README
commons Community space — posts, reactions, shared utilities README

Services

Branch Purpose Docs
api LLM access via OpenRouter (requires API key) README
backup Multi-mode backup — snapshot, versioned, Google Drive sync README
skills Capability framework for branch skills README

Back to contents


How It Works

Memory

Every branch has .trinity/ files that persist across sessions:

.trinity/passport.json       # Identity — role, purpose, principles
.trinity/local.json          # Session history — tasks, learnings, key insights
.trinity/observations.json   # Collaboration patterns observed over time

New session starts, branch reads its memories, picks up where it left off. When local files reach capacity, they roll over into @memory (ChromaDB vectors). Nothing is lost.

Standards

Every branch is held to 33 automated standards via seedgo:

drone @seedgo audit aipass              # Full system audit
drone @seedgo audit aipass @api         # Single branch

Standards cover: architecture, CLI patterns, error handling, imports, logging, naming, test quality, documentation, and more. Branches add justified bypasses in .seedgo/bypass.json.

Communication

Branches communicate via ai_mail — an internal messaging system:

drone @ai_mail dispatch @target "Subject" "Body"   # Send + wake target
drone @ai_mail email @target "Subject" "Body"      # Send without waking
drone @ai_mail inbox                               # Check your inbox

Dispatch sends a message AND wakes the target branch (starts a Claude Code session in their directory). This is how devpulse coordinates work across the system.

Structure

src/aipass/<branch>/
├── .trinity/           # Identity & memory (persists across sessions)
├── .aipass/            # Branch-specific system prompt
├── .ai_mail.local/     # Mailbox (inbox.json, sent/)
├── apps/
│   ├── <branch>.py     # Entry point
│   ├── modules/        # Business logic
│   └── handlers/       # Implementation details
├── tests/              # Branch test suite
└── README.md

All 15 branches share the same filesystem and git repo. Each owns its directory. A PR lockfile prevents concurrent git operations. Standards enforcement keeps things consistent.

Back to contents


Compliance & Safety

AIPass is built on Claude Code and operates fully within Anthropic's usage policies.

How AIPass uses Claude Code

  • Every agent session runs the official claude CLI binary (claude -p) as a genuine subprocess
  • Context is injected via Claude Code hooks (settings.json) and CLAUDE.md files — both officially supported, documented features
  • Each branch agent runs as an independent Claude Code process with its own working directory
  • No OAuth tokens are extracted, intercepted, or routed through third-party clients
  • No API calls are made to Anthropic outside the official CLI
  • Claude Code's built-in prompt caching and rate limiting are fully preserved

What AIPass does NOT do

  • No credential wrapping — we don't extract or redirect subscription OAuth tokens
  • No API proxying — we don't intercept communication between Claude Code and Anthropic's servers
  • No harness impersonation — we don't spoof the Claude Code client identity
  • No rate limit bypass — each session respects Anthropic's built-in limits

Why this matters

As of April 2026, Anthropic enforces restrictions on third-party tools that extract subscription credentials to route automated workloads outside the official CLI. Tools like OpenClaw bypass Claude Code's prompt caching optimizations, creating unsustainable compute costs.

AIPass is architecturally different: it enhances Claude Code through its own extension points (hooks, CLAUDE.md, settings.json) rather than replacing or bypassing it. Your subscription credentials stay within Anthropic's infrastructure at all times.

Using AIPass with your Claude Pro, Max, Team, or Enterprise subscription is compliant with Anthropic's terms. For server/automated deployments, API key authentication is recommended per Anthropic's guidance.

Back to contents


Project Status

Beta. Actively developed. 15 branches, 180+ PRs merged, 4,800+ tests, 100% standards compliance.

Metric Value
Branches 15
Standards 33
Tests 4,800+
PRs merged 180+
Compliance 100%
Sessions 73

For detailed progress and session history, see HERALD.md.

For per-branch status, see STATUS.md.

Back to contents


Requirements

  • Python 3.10+
  • sudo access (for global CLI symlinks during setup)
  • Claude Code (hooks provide branch identity, email notifications, auto-diagnostics)
  • API keys optional (only needed for api branch — OpenRouter/OpenAI)

Back to contents


License

MIT


Back to top

S
Description
Persistent Agent Workspace — AI agents that remember, collaborate, and never start from zero.
Readme MIT
17 MiB
Languages
Python 99.2%
Shell 0.6%
HTML 0.2%