Memcode - Agentic Coding CLI With Persistent Memory

TL;DR

TL;DR: Memcode is an agentic coding CLI that builds a persistent memory model of your codebase. Every session picks up where the last one left off instead of starting from scratch.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: memcode.ai
• Source repository: github.com/memcode-ai/cli (CLI only; install script confirms REPO="memcode-ai/cli")
• License: No license file found in repository
• HN launch thread: news.ycombinator.com/item?id=49177986

What Is Memcode?

Most coding agents start every session as a blank slate. They re-read your files, re-ask questions you already answered, and waste the first several minutes just orienting themselves. Memcode is built around the idea that an agent should remember your codebase the way a senior engineer does.

From the HN launch announcement:

“Memcode is a new developer platform currently in public beta. It includes a coding agent, chat, reusable agents, DataHub, and a Lovable-style website generator. The coding agent is heavily inspired by Claude Code and works across all frontier models as well as open-weight models such as Kimi and GLM.”

The core differentiating feature is persistent memory across sessions. Memcode builds a living picture of your codebase — how files fit together, what decisions were made, and why. When you start a new session, the agent is already oriented instead of re-learning everything from scratch.

Key Features

Persistent Codebase Memory

Memcode maintains a persistent model of your repository between sessions. Context and decisions persist, so the agent gets sharper the longer it works in your repo.

Terminal-Native TUI

A fast TUI built as a single static binary. No Electron, no browser tab, no IDE plugin required. Starts instantly and stays out of your way.

Multi-Model Support

Works across all frontier models as well as open-weight models including Kimi and GLM. The session footer shows the active model (e.g., sonnet, auto).

Reusable Agents and DataHub

Beyond a single coding agent, the platform includes reusable agent templates and a DataHub for managing structured context about your projects.

Setup

Memcode installs via a single shell command:

curl -fsSL https://memcode.ai/install.sh | sh

The installer script (verified from https://memcode.ai/install.sh) detects your OS and architecture, downloads the matching release binary from GitHub (memcode-ai/cli), verifies the checksum, and installs the memcode binary to /usr/local/bin or ~/.local/bin.

You can override defaults with environment variables:

MEMCODE_VERSION=v0.3.1     # install a specific tagged release
MEMCODE_INSTALL_DIR=~/.local/bin  # custom install location

After installation, start a session in any repository:

memcode

The terminal UI shows a status bar with version info, git status, context percentage, active model, and mode:

memcode · v0.9.3 · main · git 3✎ +120/-8 · ↑12.3k ↓4.1k · ctx 68% · sonnet · auto

How It Works

On first run in a repository, Memcode indexes the codebase to build its memory model. Subsequent sessions load that model immediately — no cold start.

The TUI offers interactive commands:

• Ask anything — query the codebase using natural language
• /plan — generate a plan for a task before executing
• /next — advance to the next logical step in a multi-step task
• Fix a bug — targeted debugging mode

The agent tracks git changes live (visible as +120/-8 in the status bar) and can work across the full history of your repository.

Deeper Analysis

Memory Model Architecture

Memcode builds a semantic understanding of your codebase rather than just indexing keywords. This is what separates it from context-window stuffing — the agent understands relationships between files and decisions.

Model Flexibility

By supporting both frontier APIs (OpenAI, Anthropic) and open-weight models (Kimi, GLM), Memcode lets teams choose between maximum capability and local/deployment flexibility. The auto mode in the status bar suggests dynamic model selection.

Comparison to Claude Code

The HN author explicitly cites Claude Code as inspiration. Memcode differentiates through persistent memory — Claude Code starts fresh each session, while Memcode compounds context over time. For large codebases with complex histories, this distinction is significant.

Practical Evaluation Checklist

• [ ] Install via curl -fsSL https://memcode.ai/install.sh | sh
• [ ] Verify binary: memcode --version
• [ ] Run memcode inside a git repository with meaningful history
• [ ] Ask a context-dependent question across two separate sessions
• [ ] Test /plan and /next commands
• [ ] Verify context loading from prior session with ctx percentage in status bar
• [ ] Test with different models (/model or env config if supported)

Security Notes

• The installer downloads binaries from GitHub releases and verifies checksums
• No cloud account required for core CLI functionality
• Memory data is stored locally in the repository (verify with ls ~/.memcode/ or equivalent post-install)

FAQ

Q: Does Memcode require an API key? A: The CLI itself does not bundle an API key. You configure your own API credentials for the model provider you choose to use.

Q: Is the codebase open source? A: The memcode-ai/cli repository exists on GitHub but has no LICENSE file and minimal README content. The product appears to be source-available with a proprietary backend. Verify licensing directly before any commercial use.

Q: How does persistent memory work across different machines? A: Memory is stored per-repository locally. If you clone the same repo on a new machine, the memory model is rebuilt from scratch rather than synced.

Q: What models are supported? A: All frontier models (Claude, GPT, Gemini) plus open-weight models including Kimi and GLM. The exact configuration depends on your API setup.

Conclusion

Memcode addresses the cold-start problem that makes every other coding agent frustrating in long-term projects. By maintaining a persistent memory model of your codebase, it eliminates the repetitive orientation tax that accumulates session after session. The terminal-native design keeps it fast and focused.

If you find yourself re-explaining your codebase to every new agent session, Memcode is worth trying. The install command is one line, and the memory benefit compounds the longer you use it.

Install: curl -fsSL https://memcode.ai/install.sh | sh