AstrivyaAstrivya

Memory for Your AI Agents

Persistent context for every AI agent — without sending your data to the cloud. Works offline. Zero configuration.

Start for free →

Context-aware code generation

// user describes feature
> Add rate limiting to API routes
Memory scan
3 related ADRs found
Generated
Your AI agent already knows your codebase architecture, naming conventions, and past decisions.
describe a feature in plain English
Astrivya surfaces relevant memories — prior implementations, related types, historical discussions.
generated code matches your existing patterns from day one
Scale

Persistent context
across every session

Your knowledge graph persists across conversations, agents, and teams. No more repeating context every time you start a new chat.

Local-first. Always available.

Your knowledge graph lives in a local SQLite file. Zero data leaves your machine unless you choose to sync. Works offline, air-gapped, or on a plane.

astrivya — zsh

One command to initialize

Run npx @astrivya/mcp init in your project. No config files, no API keys, no cloud sign-up. Your local knowledge graph is ready instantly.

utils.ts+5 −4
import { createMcpServer } from '@astrivya/mcp';
import { MemoryStore } from '@astrivya/memory';
 
export const server = createMcpServer({
name: 'astrivya-agent',
− memory: new MemoryStore({ maxEntries: 1000 }),
− tools: ['query', 'search'],
+ memory: new MemoryStore({
+ maxEntries: 5000,
+ vectorIndex: true,
+ localOnly: true,
+ }),
+ tools: ['query', 'search', 'embed', 'summarize'],
knowledgeGraph: true,
});

Query your knowledge graph

Your AI agent can search memories, recall past decisions, and reference code history — all from natural language queries.

Claude Desktop
What's the current state of the auth refactor?
From Astrivya memory: auth refactor is 70% complete. PR #142 merged (RBAC layer). Remaining: token refresh middleware. Decision log in ADR-007.
Show me the ADR summary
ADR-007: Migrate from JWT to session tokens. Rationale: better revocation, mobile support. Status: approved. Started: Feb 3.
MUniversal MCP connector

One MCP server works with every compatible agent. Connect Claude, Cursor, ChatGPT, or any tool that speaks the Model Context Protocol.

Cursor (later, same session)
Implement the token refresh middleware we decided on
From Astrivya memory: ADR-007 specifies session tokens. Middleware should intercept 401s, refresh via /auth/refresh, retry. The auth refactor PR #142 set up the base session store.
Starting implementation based on the ADR and existing store pattern...
CContext follows your agents

Memories persist across conversations and agents. Start a chat in Claude, continue in Cursor — your context is always there.

Works with
every AI.
One MCP server.

Plug Astrivya into any MCP-compatible agent — Claude Desktop, Cursor, ChatGPT, VS Code, Windsurf. Your context follows you everywhere.

Learn More ›
Scale

Memory for
your AI Agents
that persists.

Your knowledge graph stores codebase context, decisions, and history locally. Every AI agent you use can access it — zero cloud dependency.

Built for teams that need shared context

From solo developers to entire organizations. Team memory, shared knowledge graphs, and context handoffs across your engineering team.

Team Memory
Shared Knowledge Graph
Alice
Alice
Bob
Bob
Carol
Carol
Dave
Dave

Team memory that works together

Share knowledge graphs across your team. Every member's AI agent has access to shared context — decisions, architecture, and project history.

Knowledge Graph
Claude
Cursor
ChatGPT
VS Code

Works with any MCP agent

Claude Desktop, Cursor, ChatGPT, VS Code — any MCP-compatible agent can tap into your team's shared knowledge graph.

Install
Initializing Astrivya MCP server...
✓ Knowledge graph created
✓ Server running on port 3000
Sources

Run one command

npx @astrivya/mcp init in your project. No config files, no API keys, no cloud sign-up. Your local knowledge graph is ready.

Connect
astrivya-mcp start --port 3000
MCP server connected to Claude Desktop
4 tools available: query, search, embed, summarize
Sources

Point your agent to the server

Connect Claude Desktop, Cursor, ChatGPT, or any MCP-compatible agent. One-line setup, works with every tool.

Query
Why did we choose SQLite over Postgres?
From memory: decision logged Mar 12 — local-first requirement, offline support, zero ops. ADR-004 has full rationale.
Sources

Ask anything with full context

Your AI agent now has persistent access to your codebase history, decisions, TODOs, and architecture. Context survives across sessions.

Build more. Pay less. Scale smart.

Try free for 14 days with full access to all features. Upgrade only when your AI agent proves value.

Starter

$0

For solo developers getting started

  • Personal agent & memory
  • 3 active connectors
  • 1,000 active memories
  • 500K monthly token allowance
  • Community support
  • MCP server access

ProPopular

$19

For professionals who need more power

  • Personal agent & memory
  • 5 active connectors
  • 5,000 active memories
  • 5M monthly token allowance
  • Daily briefings & Slack bot
  • 5 custom skill slots
  • 50 workflow runs / month
  • Priority support
Full feature comparison →

Frequently asked questions

Get answers to commonly asked questions.

Your AI agents deserve
persistent memory

Give every AI agent in your toolkit shared context across sessions. One command, zero configuration, free forever.

Install Astrivya