Skip to content
NarraNexus
Multi-Agent Product · Open Source

Agents that remember, connect, and work as one.

You are shaped by what you've lived: the experiences you carry, the people you've met, the goals that slowly became yours. Agents start without that history. But when an agent stays with you, time begins to matter.

Narra gives agents memory over time, organized into living storylines. Nexus gives them a network of people, tools, and other agents. Together, they become long-lived collaborators: agents that remember your context, understand your relationships, build a working model of your world, and coordinate as a team to get things done.

Try Onlinew/ free token quota
DownloadDocsGitHub

Run the cloud version, the macOS desktop app, or the local stack from source — the same agents, three surfaces.

Narra · Narrative memory · it understands the story

It remembers you, and understands your story.

Automation that forgets you isn't help, it's homework. Narra is the memory that makes an agent feel like it actually knows you.

Picks up where you left off

Tuesday's thread, continued Monday. Memory is organized by topic, not timestamp, so weeks later your agent already knows what you meant. No recap.

Holds the long arc

Past the current chat, a long-term memory keeps the throughline of a project, a goal, a relationship. Nothing important resets.

In your voice

It learns your tone and your patterns: the companion that remembers what you told it last Tuesday, and sounds like it.

Yours, and private

Your memory stays yours. Run it locally, bring your own keys, keep every conversation to yourself.

oldernow
Lifespan by topicretrieved by meaning, not recency
agent.narra.nexus / memory
Incoming
“Any update on the lead investor’s term sheet?”→ Q3 FUNDRAISING
Storylines4 active
Q3 fundraising14 events2 days ago
Competitor research31 events3 weeks ago
Hiring — design9 events6 days ago
Product roadmap22 eventsyesterday

Retrieved by meaning, not recency. No shrinking context window.

Nexus · Nexus network · it connects, and gets it done

It connects, and gets it done.

An agent alone is a tool. Connect it (to other agents, to your people, to your tools) and it becomes a team that does the work.

A team, not a tool

Ask once. A coordinator hands each piece to the right specialist and brings back one answer. Agents talk to each other, hand off, and coordinate. You manage the outcome, not the steps.

Knows your people

Every person it meets joins a graph it maintains: role, style, history, how close. It talks to each of them appropriately, across Slack, Telegram, Lark, and WhatsApp.

Works while you sleep

Schedule it once. Briefings, triage, and reports run in the background and ping you only when they're done.

Reaches the whole world

Files, the web, your apps, thousands of skills. Describe what you need and your agent installs the tool itself. No wiring.

agent.narra.nexus / jobs
Jobs· 14
ListGraphTimeline
Active21
Success1213
Failed0
Rate100%
Status Distribution14 total
1213completed21active
AllActiveRunningCompletedFailed
Lark Daily DigestCompleted
Pushed yesterday’s highlights at 09:00 · 4 channels · 2 attachments
Morning Briefing ReportActiveCompleted
Briefing Maestro coordinating 6 desk analysts · drafting headlines
job_a3f4
Feature Dev · OAuth 2.1 + PKCEActive
Auth squad migrating callbacks · 12 of 18 endpoints rewired · tests passing
job_c92e

Jobs & long tasks · works while you sleep

Schedule one-offs, cron jobs, or always-on daemons. Agents work in the background and report back through chat when the results are ready.

Learn more
agent.narra.nexus / team
Agents5
SAStrategy Advisor12
MAMarketing Agent8
SLSales Agent4
FIFinance Agent2
COContent Writer0
+ ADD AGENT
Interactionagent_de9d27
PROCESSING
SA
STRATEGY ADVISOR
Q4 board prep, M&A review, growth targets. Top priority: close the two pending contracts before Friday.
MA
MARKETING AGENT
Weekly post draft is queued for your review.
Publish it, push the contracts.
ME
Ask the team⌘↵

Team collaboration · DM, group chat, hand-offs

Spin up a research desk, a content team, or an on-call rotation. Agents coordinate over a shared message bus: DM, group chat, hand-offs, and shared context across the whole team.

Learn more
agent.narra.nexus / awareness
IM Channels3 / 5 bound
Lark / Feishu@narra.foundation✓CONNECTED
Slacknetmind-internal · #agents✓CONNECTED
Telegram@narra_briefings✓CONNECTED
People23
Tao Wang
Person·32 chats·Direct
STRONG
Marcus Klein
Person·18 chats·Direct
MEDIUM
Sora Nakamura
Person·1 chat·Known of
WEAK

Social network · channels · contacts

Every person, team, and channel it meets joins a graph your agent maintains: roles, expertise, communication style, history. Lives on Discord, Slack, Telegram, and WhatsApp out of the box.

Learn more
Narra + Nexus

Most agent frameworks organize agents. NarraNexus organizes your worlds.

You sit at the center of two: your life on one side, your work on the other. Each is a living graph of the people, tools, and agents behind it, kept in order by the same memory and network. One place, both worlds.

PERSONALPROFESSIONALYouHealth AgentFinance AgentFamily AgentResearch DeskGrowth AgentEngineering StudioMichaelMarket APIAnalyst AgentGitHubVercelDeveloper Agent…
Teams

Browse and import. One click.

Ready-made teams: research desks, content studios, development pipelines, briefing bots. Pick the one closest to what you need and it boots with its agents, skills, and shared memory intact.

Browse all teams
Teams18
Financial Morning Briefing
6 agents
Marketing Team
4 agents
Web Development Team
4 agents
ME
YOU06:30 GMT
Give me today's brief.
BM
BRIEFING MAESTRO
On it. Loading Daily Brief narrative, querying 5 analysts.
BM
5 / 5 responses in. Drafting brief and formatting tables.
BM
✦ Writing morning-brief.html
BM
✓ Brief is ready. Asia closed flat; Fed dot plot at 14:00 ET.
morning-brief.html
morning-brief.htmlLIVE
Jun 9 · 06:30 GMT
Asia closes flat. Watch the Fed dot plot at 14:00 ET.
Nikkei ↑ 0.8%S&P futures ↑ 0.2%NVDA +1.4%
VIEW FULL BRIEF →
ME
YOU09:00 GMT
Process today's sponsor pipeline.
SM
SPONSORSHIP WORKFLOW MGR
On it. Scanning inbox and brand mentions.
SM
Email Intake parsed 3 sponsor emails. ReviewScout tracked 12 mentions.
SM
✦ CRM Agent updating deals. Writing daily summary.
SM
✓ Pipeline updated. 2 deals advanced to negotiation.
pipeline-summary
Email sentgmail.com
Subject
Tonight at 6: live walkthrough of our agent engine
1,247
recipients
42%
open rate
8%
click rate
VIEW CAMPAIGN →
ME
YOU14:08 GMT
Ship the coffee landing page.
PM
PRODUCT MANAGER
On it. Briefing Web Developer + Design Reviewer.
PM
Web Developer built site. Reviewer flagged 3 small fixes, applied.
PM
✦ Vercel Deployment Agent shipping the build.
PM
✓ Site is live at tonight-roast.vercel.app. Lighthouse 98.
deploy.log
tonight-roast.vercel.app
Tonight RoastMENU · ABOUT · SUBSCRIBE
Tonight’s roast,
today.
Single-origin beans, roasted at 4 PM, brewed in your kitchen by 8.
Start your subscription →
DEPLOYED · 3m AGOLIGHTHOUSE 98 / 100
OPEN APP →
Get started

A working agent team in three steps.

No config sprawl. No bespoke wiring. Pick a surface, drop in a team, start the conversation.

Step 01

Set it up

Open the cloud app, install the macOS build, or run it locally from source. Pick the surface that fits. Your team is ready in minutes.

WEB · DESKTOP · LOCAL
Step 02

Choose a team

Pick a ready-made agent team: a research desk, a content studio, a briefing bot. Or start blank and shape your own.

TEAMS
Step 03

Import & start talking

One click to import. Your agents are online, share memory, and use tools out of the box. No wiring, no glue code.

ZERO CONFIG
Open, local, yours

Open, local, yours.

Try it end to end before you bring anything. Nothing hidden, nothing locked in.

Open source

Read every line. Nothing hidden, nothing to trust on faith.

Runs anywhere

Cloud, a macOS app, or fully local: the same agents, your choice of surface.

Your keys, your models

Claude, OpenAI, Gemini, or your own endpoint. Bring what you already pay for.

Free to start

A token quota to try the whole product before you bring anything of your own.

FAQ

Common questions

Is there a free tier?
Yes. New cloud accounts get a system-provided token quota covering Agent inference, embeddings, and helper calls. Enough to try the product end-to-end before you bring your own keys.
Can I bring my own model or API key?
Yes. NarraNexus runs on a three-slot architecture: Agent, Embedding, and Helper. You can plug in NetMind.AI Power (covers all three), OpenRouter, Yunwu, or any custom Anthropic / OpenAI-compatible endpoint.
Do my chats stay private?
Yes. Every agent is private by default: only you see its conversations.
Will my agent remember things between sessions?
Short-term memory in the active conversation is handled by the Chat module. Cross-session long-term memory lives in Narrative. Your agent picks up the right thread even weeks later, organized by topic rather than timestamp.
Can I run it fully locally?
Yes. The macOS desktop app ships with a bundled runtime and runs entirely on your machine: no account, no network round-trip. From source, one bash run.sh starts the full stack.
Multi-device login?
Supported. JWT tokens are per-device, so you can stay signed in on laptop, desktop, and phone in parallel. Agent narratives live server-side and stay in sync across them.

Full beginner guide and FAQ→

Roadmap

Where we are going

  1. 2026-03
    v1.0FoundationShipped
    • Core agent runtime with 7-step pipeline
    • Narrative memory engine
    • Module system with hot-pluggable capabilities
    • Built-in MCP tools
    • Multi-LLM provider support
  2. 2026-03
    v1.0.5Desktop & CloudShipped
    • macOS desktop app with bundled runtime
    • Cloud deployment with multi-tenant auth
    • Agent-to-agent communication (DM & group chat)
    • Skill system & ClawHub marketplace integration
    • Dashboard v2 with job tracking & cost monitoring
  3. 2026 Q2
    v2.0Scale & Integrate
    • Docker Compose one-command deployment
    • IM platform integrations (Telegram, Lark)
    • Community plugin system
    • Multimodal interaction support
Tonight

Talk to your first agent tonight.

Sixty seconds from open to your first reply.

Try Onlinew/ free token quota
DownloadDocsGitHub