ROUTING
Cross-Tool Interoperability
Bridge CLI agents (Claude Code, Aider), IDE assistants (Cursor, VS Code), and custom scripts over a unified peer-to-peer and relayed event mesh.
Universal Protocol & Live Shared Memory for AI Agents
The layer between
thought and action.
june.stream connects coding agents across tools, runtimes, and machines. An open, real-time communication protocol paired with durable shared memory infrastructure—so intelligent agents can collaborate without losing context.
OPEN PROTOCOL
SHARED MEMORY INFRASTRUCTURE
BUILT FOR AGENT WORKFLOWS
/ the idea
Work should have
a continuous memory.
Today, every coding agent is an isolated island. Your terminal CLI agent (Claude Code, Aider) knows nothing about what your IDE assistant (Cursor, Windsurf) is editing. When work hands off to a cloud sandbox or CI/CD runner, all reasoning context evaporates.
Every session begins from zero: re-ingesting repository tokens, re-discovering file structures, and blindly stepping on concurrent changes.
june.stream is the connective tissue: an open, universal communication protocol and live shared memory fabric that lets autonomous coding agents coordinate across tools, machines, and time.
PRODUCT 01 / CONNECTIVITY
Universal Communication Protocol
connecting agents across tools and machines.
A standardized, lightweight wire protocol for multi-agent software engineering. Seamlessly orchestrate coding agents running in local terminals, desktop IDEs, and remote cloud runtimes over encrypted peer-to-peer and pub-sub channels.
STREAMING
Streaming Tool Envelopes
Bidirectional JSON envelopes for streaming file patches, diagnostics, subagent task delegation, and approval gates.
TUNNELING
Machine-to-Machine Mesh
Encrypted tunnels linking local developer workstations with remote execution sandboxes (Modal, E2B, Daytona, Docker containers).
CONSENSUS
Deterministic Swarm Consensus
Subagent locking mechanisms and conflict-free execution ordering preventing duplicate writes across parallel agent workers.
TRANSPORT SPECIFICATION
- Wire Transports: WebSockets (TLS 1.3), Unix Domain Sockets, QUIC, WebRTC DataChannels
- Framing: Streaming binary envelopes + JSON-RPC 2.0 schema
- Security: Mutual TLS & Noise Protocol Framework end-to-end encryption
- Addressing: URI scheme
june://<channel>/<agent-id>
AGENT RUNTIMES WE'RE BUILDING ADAPTERS FOR
PRODUCT 02 / CONTINUOUS CONTEXT
Live Shared Context & Memory
infrastructure for autonomous AI agents.
Stop re-prompting. june.stream provides distributed working memory that maintains live codebase state, architectural choices, and runtime outputs across sessions and machines — synchronized by live deltas, not full re-ingestion.
LIVE SYNC
Live State Deltas
Continuous memory delta synchronization ensures all collaborating agents observe identical state in real time.
DAG FABRIC
Durable Context Graph
A structured, queryable context DAG tracking active plans, architectural constraints, modified symbols, and past verification runs.
PORTABILITY
Instant Session Resurrection
Pause work on your local machine and resume it hours later in a remote cloud container without losing reasoning traces or scratchpads.
EFFICIENCY
Token Compression Engine
Shared reference pointers and delta diffing cut redundant prompt bloat — peers read live memory instead of re-ingesting context.
✕ WITHOUT JUNE.STREAM (ISOLATED AGENTS)
- Each tool runs in a solitary walled garden with zero knowledge of peer agents.
- Context resets completely when terminal closes or cloud runner shuts down.
- Redundant token consumption: every agent re-reads entire codebases on launch.
- Disjointed manual handoffs: developer must manually copy diffs and error logs.
- Race conditions and duplicate file edits when running multiple agents simultaneously.
✓ WITH JUNE.STREAM (CONNECTED AGENT MESH)
- Unified peer mesh: Cursor, Claude Code, and Cloud runners share live state.
- Durable session resurrection: pick up work on any machine right where you left off.
- Shared context by reference: peers read live memory instead of re-reading the whole codebase.
- Instant automatic handoff: local agent delegates test execution to cloud sandbox.
- Deterministic conflict resolution and lock management across agent swarms.
/ 03 architecture
Distributed Topology
for the agent mesh.
The topology we're building toward: heterogeneous developer environments joined into one synchronous mesh.
LOCAL EDGE NODES
Cursor IDE Plugin Source Editor & Diffs
Claude Code (CLI) Terminal Orchestrator
Local Host Daemon IPC / Domain Sockets
↕ TLS 1.3 / mTLS Mesh ↕
JUNE.STREAM CORE FABRIC
Universal Message Router JSON-RPC & Event PubSub
Distributed Context Graph (DAG) Working Memory & Symbol State
Locking & Swarm Consensus Conflict-Free Delta Resolution
↕ WireGuard / WebSockets ↕
CLOUD & RUNTIME NODES
E2B / Modal Sandbox Headless Test Execution
CI/CD Agent Runners Automated Build & Lint
Durable Storage Sync Cloudflare DO / SQLite
/ 04 developer experience
The reference client,
reading like the protocol.
june.stream ships a small TypeScript reference client implementing the protocol primitives: typed envelopes, presence heartbeats, last-write-wins shared memory, and ping/pong round-trip measurement. The examples below mirror its public API.
- ✓ Runs in any modern browser — no install, no backend
- ✓ Tab-to-tab envelopes over the BroadcastChannel API
- ✓ Live presence, typed events, measured peer latency
- ✓ Shared memory persisted with localStorage — survives reloads
import { JuneStream } from './june/stream';
// 1. Join the mesh — pick any channel, name your agent
const stream = await JuneStream.connect({
channel: 'project-altair',
agent: 'cursor-agent'
});
// 2. React to every envelope on the mesh in real time
stream.on('*', (envelope, receivedAt, echo) => {
console.log(envelope.type, 'from', envelope.from, echo ? '(you)' : '');
});
// 3. Dispatch work to a peer — or '*' for every connected agent
await stream.dispatch({
target: 'cloud-runner',
action: 'eval_suite',
payload: { files: ['src/auth.ts'] }
});INVESTOR PRESENTATION
june.stream Seed Pitch Deck
Review our market thesis, technical architecture, and product roadmap. Formatted as an editable, 12-slide PowerPoint presentation built directly from this repository.
Make the distance between
now and next disappear.
Building agent workflows? Tell us what you're working on — the team reads every message.