Contxt — Memory for AI Agents

~/my-saas-app

$contxt init

✓ Initialized in ./my-saas-app

Platforms: claude-code · gemini · vscode-copilot

$contxt push

✓ Synced 14 decisions · 8 patterns

$contxt sessions resume

✓ Context restored from last session

Goal: payment integration · 3 tasks in progress

$contxt stats

62% token reduction · $0.84 saved

Context always available

Memory that persists
across every session

Your architecture. Your decisions. Always there.

Every AI coding session starts from zero. You spend 40% of your prompts re-explaining your architecture, your patterns, your past decisions. Your AI doesn't remember that you use Prisma, that you prefer server actions over API routes, or that you already decided against Redis.

Worse — when the context window fills up and your AI compacts the conversation, all that momentum is gone. Hours of session history, decisions in progress, blockers you were working through. Vanished.

Contxt solves this. Persistent project memory, rolling session snapshots, token analytics, and zero-config setup for every AI tool you use.

40%

of prompts wasted on re-explaining context

3.2×

faster iteration with persistent memory

5K+

tokens saved per session on average

manual copy-paste. Contxt loads automatically.

#01 Workflow

Push once.
Never repeat yourself.

Store decisions, patterns, and active context in your project's .contxt/ directory. One command syncs it to the cloud. Every AI tool you use pulls from the same source of truth.

✓Decisions (why you chose Prisma over TypeORM)
✓Patterns (your API error handler template)
✓Context (current blockers, active features)
✓Documents (API references, onboarding guides)

Push to cloud

Sync happens automatically

~/my-saas-app

$contxt init --platforms claude-code,gemini,vscode-copilot

✓ .mcp.json (Claude Code)

✓ ~/.gemini/settings.json

✓ .vscode/mcp.json (Copilot)

✓ CLAUDE.md (agent instructions)

✓ GEMINI.md

All platforms configured in one command.

#02 Integrations

Every AI tool.
One command to rule them.

Contxt uses MCP — the open standard for AI context. Run contxt init once and it detects which AI tools you have installed, writes the right config for each, and injects agent instructions so they know exactly how to use your memory.

Claude CodeGemini CLIVS Code CopilotCursorWindsurfOpenCodeCodex CLI

#03 Structure

Git-like branching.
Time travel included.

Want to experiment with a new approach? Create a branch. Try it out. Merge it back or throw it away. Every change is versioned — you can revert to any point in your project's memory history.

$contxt branch create experiment

$contxt decision add --title "Try approach B"

$contxt branch merge experiment

main

Decision: Use Prisma

Pattern: API errors

experiment

Decision: Try approach B

The relevance engine

Contxt doesn't dump everything into your prompt. It intelligently scores each memory entry and surfaces only what matters for your current task.

Task Signal

QUERY

"implement Stripe webhook handler"

ACTIVE FILES

src/app/api/webhooks/route.ts

src/lib/stripe.ts

TOKEN BUDGET

max_tokens: 2000

Ranked Results

5 entries • 1,847 tokens

Decision: Use Stripe for payments

0.94

Keyword overlap • File match • Type priority

432 tokens

Pattern: API error handler

0.87

Recency • Type priority

318 tokens

Context: Payment flow blockers

0.82

Semantic similarity • Recency

245 tokens

Filtered 18 irrelevant entries • Saved 4,203 tokens

#04 Session Continuity

Context resets.
Your momentum doesn't.

Contxt logs every notable event during your session — decisions made, files edited, tasks completed, errors hit. Every 10 events it builds a rolling snapshot. When Claude hits the context limit, one command restores everything.

1Your AI silently logs events via contxt_session_event — no manual work required
2A compact snapshot is rebuilt automatically every 10 events and stored in the cloud
3Start a new session, run contxt sessions resume — full context restored in one tool call

claude code — 3h 12m

⚠ This conversation cannot be compacted anymore.

Start a new session to continue.

~/my-saas-app — new session

$contxt sessions resume

✓ Context restored

Goal: payment integration · 47 events logged

Completed: Stripe handler, signature middleware

In progress: retry queue · Next: integration tests

#05 Analytics

Know your memory health

Track token savings, see what changed since your last session, and surface stale entries before they mislead your AI.

contxt stats

📊 my-app · last 30 days

MEMORY

decisions 12

patterns 8

total 23

TOKEN EFFICIENCY

avg reduction 62%

est. cost saved $0.84

MOST RETRIEVED

1. JWT in httpOnly cookies — 18×

2. API Error Handler — 14×

contxt diff

Context diff — since 2h ago

+[decision] Switch to Edge runtime

+[pattern] Optimistic UI updates

~[decision] Use Prisma for ORM v3→v4

~[pattern] API Error Handler v2→v3

?[decision] Use Redux — stale · 45d

2 new · 2 updated · 1 stale

Token efficiency

Track avg tokens returned, reduction %, and estimated cost savings per project.

Context diff

See exactly what changed since your last session — new, updated, and stale entries at a glance.

Stale detection

Surface memory that hasn't been updated in 30+ days before it starts misleading your AI.

One integration.
Every tool.

Contxt uses MCP, the open standard for AI context. Run contxt init once — it detects installed tools and configures them all.

Claude Code

Gemini CLI

VS Code Copilot

Cursor

Windsurf

OpenCode

Codex CLI

Zed

View MCP setup guide →

Simple, transparent pricing

Start free. Scale as you grow.

Free

Forever free

✓3 projects
✓Up to 1,000 memory entries
✓Local + cloud sync
✓MCP integration (all platforms)
✓Session continuity
✓Basic search

Get started

Memory that survives
every session, every
context reset

Memory that persists
across every session

Push once.
Never repeat yourself.

Every AI tool.
One command to rule them.

Git-like branching.
Time travel included.

The relevance engine

Context resets.
Your momentum doesn't.

Know your memory health

One integration.
Every tool.

Simple, transparent pricing

Give your AI
a memory

Memory that survivesevery session, everycontext reset

Memory that persistsacross every session

Push once.Never repeat yourself.

Every AI tool.One command to rule them.

Git-like branching.Time travel included.

The relevance engine

Context resets.Your momentum doesn't.

Know your memory health

One integration.Every tool.

Simple, transparent pricing

Give your AIa memory

Memory that survives
every session, every
context reset

Memory that persists
across every session

Push once.
Never repeat yourself.

Every AI tool.
One command to rule them.

Git-like branching.
Time travel included.

Context resets.
Your momentum doesn't.

One integration.
Every tool.

Give your AI
a memory