Memcode 发布终端编码代理，为代码库构建持久记忆模型

memcode · terminal

↺ ready main · sonnet · auto · aurora · mirror · extra mile● session→ memcode · v0.9.3 · main · git 3✎ +120/-8 · ↑12.3k ↓4.1k · ctx 68% · sonnet · autoAsk anything/plan/nextFix a bugMemcode · agentic coding with memory

Your repo has a history.
Your agent should know it.

Memcode remembers your codebase the way a senior engineer does, then puts a coding agent on top of that memory, right in your terminal. Install in one line.

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

Prefer to inspect first? Read the install script

Why Memcode

Not another chat wrapper. A persistent memory system that makes the agent actually useful, because it knows what came before.

Knows your repo, not just your files

Memcode builds a living picture of your codebase: how it fits together, what was decided, and why. Every session starts already oriented instead of re-reading everything.

Never starts from zero

Other agents burn the first ten minutes rediscovering your codebase and still miss the why. Memcode picks up exactly where it left off, without spending your tokens to re-learn.

Lives in your terminal

A fast TUI in a single static binary. No Electron, no browser tab, no IDE plugin. It starts instantly and stays out of your way.

Memory that compounds

Context and decisions persist across sessions. The agent gets sharper the longer it works in your repo, because it remembers what it learned last time.

The first session is the last cold start.
Every session after starts ahead.

Every other agent starts from scratch each session, re-reading files and re-asking questions you already answered. Memcode remembers, and it compounds.

01

Work like you always do

Commit, edit, ship. No annotations and no special workflow. Memcode watches your repo and keeps up on its own.

02

It remembers what matters

Not just the code, but the decisions, the dead ends, and the reasons. The context you would otherwise re-explain every session.

03

Every session starts ahead

Open the terminal and the agent already knows where things stand. No warm-up, no re-reading, no "let me look around first."

Chat, wired to your knowledge

Every frontier model in one place, with your DataHub docs on tap. Ask a question, get an answer grounded in your team's documents, and share the thread.

mMemcodeNew chatAgentsAssistantsDataHubWebsites

Recent

Refactor auth middlewareDebug race condition in workerReview PR #142 streaming fixPlan migration to new schemayou@memcode.aiNew chat· 2 docs in contextpayments-overview.mdWhat is our refund policy for EU customers on annual plans?

Here is what your docs say.

The policy

- Annual plans get a full refund inside the first 30 days, no questions asked

- After 30 days, refunds are prorated to the unused months

- EU customers are always refunded through the original payment provider

Worth knowing

- The 30-day window is a legal requirement in the EU, not just policy

- Prorated refunds need a support ticket today; self-serve is on the roadmap

Source

- payments-overview.md, refunds section, updated in April

2 docs

Agents that work while you don't

Describe the job in plain language and get a custom agent: research, reports, monitoring, follow-ups. Run it on demand, on a schedule, or from a webhook, with web search, email, and your connected tools.

mMemcodeNew chatAgentsAssistantsDataHubWebsites

Recent

Refactor auth middlewareDebug race condition in workerReview PR #142 streaming fixPlan migration to new schemayou@memcode.ai

Competitor Watch Agent

Scheduled · Mondays at 9am

RunningConnected:Web SearchEmailStorage

Starting the weekly competitor sweep...

2s

Searching the web for launches and pricing changes...

14s

Reading 12 announcements and changelogs...

9s

Writing this week's brief...

8s

Emailing the brief to the team...

Output Preview

Weekly brief: 3 findings

Two competitors moved to usage-based pricing this week.

One shipped an enterprise tier. Full comparison saved to storage and emailed to the team.

Give chat your team's knowledge

Architecture decisions, runbooks, API references, design docs. Upload them to DataHub and chat on the web can search them and cite them in its answers.

mMemcodeNew chatAgentsAssistantsDataHubWebsites

Recent

Refactor auth middlewareDebug race condition in workerReview PR #142 streaming fixPlan migration to new schemayou@memcode.aiDocsUploadConnections

Name	Size
Architecture Decisions18 items	—
Runbooks6 items	—
Product Specs11 items	—
api-reference.md	28 KB
payments-overview.md	12 KB
deploy-runbook.md	15 KB
onboarding.md	6 KB
schema.sql	9 KB

Describe a site. Ship it.

The website builder turns a prompt into a real site: watch it build live, click any text to edit it in place, then publish to your-name.memcode.app or bring a custom domain.

mMemcodeNew chatAgentsAssistantsDataHubWebsitesyou@memcode.aiBuild a landing page for a specialty coffee roaster. Warm, minimal, with a story section and an order CTA. Build

Built the landing page: hero with tasting notes, a story section, and an order CTA wired to the menu.

Add a subscription card with three tiers

Working on your site…

Describe a change to build…ember-coffee.memcode.appEmber RoastersBeansStoryVisitOrder

Small-batch since 2019

Coffee roasted the slow way

Single-origin lots, roasted to order and shipped within 24 hours of the roast.

Order beansOur story

Ethiopia Guji

peach · bergamot

Huila Decaf

cocoa · plum

Ember Blend

toffee · citrus

Every frontier model

Different models think differently. Memcode sends each job to the model that handles it best, or you pin one and it handles the whole session. Direct-billed at provider cost, or bring your own keys.

O

OpenAI

3 models

GPT-5.6 Sol

thinking

1.1M context

OpenAI frontier GPT-5.6 Sol. Top-tier intelligence for coding and professional work, with a 1.05M context window.

GPT-5.6 Terra

thinking

1.1M context

Balanced GPT-5.6 model with strong reasoning at a fraction of the cost of Sol.

GPT-5.6 Luna

thinking

1.1M context

Fast, cost-effective GPT-5.6 model for quick tasks and high-volume use cases.

A

Anthropic

4 models

Fable 5

thinking

1M context

Anthropic's most capable model: state-of-the-art coding, vision, and computer use with a 1M context window and always-on adaptive thinking.

Opus 5

thinking

1M context

Exceptional Claude model for advanced software engineering, with a 1M context window and adaptive thinking.

Sonnet 5

thinking

1M context

Latest Claude Sonnet with adaptive thinking, 1M context, and best speed-to-intelligence ratio.

Haiku 4.5

200K context

Fastest and most compact Claude model, perfect for quick responses.

G

Google

2 models

Gemini 3.1 Pro

thinking

1M context

Latest Gemini model with significantly improved reasoning and 1M context window.

Gemini 3.6 Flash

thinking

1M context

Gemini 3.6 Flash, the latest frontier speed model. Combines frontier intelligence with superior search and grounding; outperforms Gemini 3.1 Pro on coding and agentic benchmarks at a fraction of the cost.

X

xAI

1 model

Grok 4.5

thinking

500K context

xAI flagship model with large context window.

F

Fireworks

4 models

GLM-5.2

thinking

1M context

Open-weight coding flagship (Z.ai) with a 1M context window — the same cheap lane the memcode CLI runs on.

Kimi K2.6

thinking

262K context

Moonshot general open-weight model with native vision and a 262K context window.

Kimi K2.7 Code

thinking

262K context

Moonshot coding specialist with native vision and a 262K context window.

Kimi K3

thinking

1M context

Moonshot flagship on the cheap Fireworks lane: 1M context and native vision.

Install in one line

macOS, Linux, and Windows. Runs in your terminal. Run on Memcode credits, or bring your own keys with any plan.

$curl -fsSL https://memcode.ai/install.sh | shLog in to the web app