Noureddine RAMDI / agentic-stack: portable multi-agent memory for AI coding assistants

Created Tue, 05 May 2026 13:37:39 +0000 Modified Sat, 23 May 2026 20:41:27 +0000
912 Words
github-stars python ai-agents agent-memory multi-agent cli cross-platform

codejunkie99/agentic-stack

agentic-stack addresses a persistent pain point for developers working with multiple AI coding agents: the loss of agent memory and skills when switching between different harnesses. If you’ve ever toggled between Claude Code, Cursor, or other AI coding assistants, you know how frustrating it is to lose project context, preferences, or episodic histories. This repo offers a unified, portable solution to keep agent memories intact across various tools.

what agentic-stack does and how it is built

At its core, agentic-stack provides a portable .agent/ folder that acts as a shared memory-and-skills layer spanning multiple AI coding agents. It supports a broad range of harnesses including Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Hermes, Pi, Codex, Antigravity, and DIY Python agents.

The main architectural elements include:

  • Manifest-driven adapter system: Each harness has an adapter.json manifest describing how to interface with its memory and skills. This abstraction enables seamless switching and integration.

  • Local data layer and telemetry: The repo tracks token use, cost estimates, KPI summaries, and exports dashboards locally. This local-first approach allows developers to monitor multi-agent suites without relying on remote services.

  • Data flywheel: Approved runs are exported into trace records, context cards, evaluation cases, and training-ready JSONL datasets. This supports continuous improvement and auditing.

  • Transfer wizard: A natural-language onboarding wizard enables migrating project brains between harnesses with SHA-256 verification for integrity. It auto-detects target harnesses, memory scopes, and generates curl or PowerShell bootstrap commands.

The repo is primarily written in Python, designed to be cross-platform compatible on POSIX systems and Windows. Installation options include Homebrew taps or native installers, catering to different developer environments.

technical strengths and tradeoffs

What distinguishes agentic-stack is its harness-agnostic approach to agent memory management. Instead of tying memory to a single tool or vendor, it externalizes it into a portable folder with a well-defined manifest interface. This tackles a real problem: agent memory lock-in that complicates switching or combining AI coding assistants.

The manifest-driven adapter system is a clean, modular design that isolates harness-specific quirks behind JSON manifests. This makes adding or updating adapters manageable, though it does mean maintaining these manifests and keeping them in sync with evolving harness APIs.

The local telemetry and data flywheel provide rich insights into usage and performance. This local-first model avoids cloud dependencies and offers granular control, though it introduces additional complexity and resource use on the developer’s machine.

The transfer wizard is the standout feature, blending natural-language parsing with strict integrity verification. It generates ready-to-run commands for bootstrap migration, which is a thoughtful solution to a subtle but impactful pain point.

On the downside, the complexity of supporting many harnesses and the overhead of telemetry may be overkill for smaller projects or single-agent setups. The Python codebase appears clean and well-structured but will require maintenance effort as new AI agents emerge.

quickstart

macOS / Linux

# tap + install (one-time — both lines required)
brew tap codejunkie99/agentic-stack https://github.com/codejunkie99/agentic-stack
brew install agentic-stack

# clone + run the native installer
git clone https://github.com/codejunkie99/agentic-stack.git
cd agentic-stack
.\install.ps1 claude-code C:\path\to\your-project

Already installed?

brew update && brew upgrade agentic-stack

Clone instead?

git clone https://github.com/codejunkie99/agentic-stack.git
cd agentic-stack && ./install.sh claude-code         # mac / linux / git-bash

### Already installed?

```bash
brew update && brew upgrade agentic-stack

or on Windows PowerShell: .\install.ps1 claude-code

Once installed: manage what’s wired

After the first ./install.sh, manage your project with verb-style subcommands:

./install.sh add cursor          # add a second adapter (Claude Code + Cursor in same repo)
./install.sh status              # one-screen view: which adapters, brain stats
./install.sh doctor              # read-only audit; green / yellow / red per adapter
./install.sh manage              # interactive TUI: header pane + menu loop for add/remove/audit
./install.sh transfer            # onboarding-style wizard: export/import memory as a curl bridge
./install.sh remove cursor       # confirm prompt + delete; no quarantine, no undo

Bare ./install.sh (no arguments) opens a multi-select wizard on a fresh project, auto-detecting harnesses already on disk and pre-checking them.

verdict

agentic-stack is a solid choice if you juggle multiple AI coding agents and need a consistent, portable way to preserve and migrate agent memory and skills. Its manifest-driven adapters and local telemetry provide transparency and control, while the transfer wizard solves the often overlooked problem of memory lock-in between AI tools.

It is less suited if you only use a single agent or prefer a cloud-based solution, as the local data layer adds complexity and resource use. Maintenance of adapter manifests requires some upkeep, especially as AI harnesses evolve.

Overall, the repo is a practical tool that fills a niche in AI agent workflows with a clear, well-implemented architecture and developer-friendly management commands.

→ GitHub Repo: codejunkie99/agentic-stack ⭐ 1,835 · Python