MESH ONLINE│CODENAME: Start Me Up
NEWATHENA AGENTS 1.0 · DESKTOP APP

your agents. one station.

A home for your AI agents. Build teams, give each agent a job, a model and the tools it needs, and watch them work together on a live map.

No account and no login: it runs on your computer. Works with Claude Code, Codex, opencode, local models and MCP. Free and open source.

windows · macos · linuxmit or apache-2.0
§01 / the station

every agent. on one live map.

The Station is your project as a live 3D map. You see who is working, who is waiting on you, and which tools they are using, without opening a single tab.

the station, three teamsdemo data
The Station: the Website, Launch and Research teams around the Brain, with the programs they use along the top
the station, three teams
1

Teams are pods

Each team is a pod with its agents at their desks. Click a team to open its panel and chat.

2

The Brain sits in the middle

It runs the project for you: it sets up teams, hands out the work and checks it got done.

3

Lines light up when work happens

Programs and services run along the top. A line lights up each time an agent calls one, and the feed at the bottom left shows what's happening now.

4

Pick a look

Turn on Voxel world and your agents build a little world around the teams, or Scene: a web page your agents draw for the project, live.

§02 / the brain

describe the job. get a team.

The Brain is the assistant that runs the project for you. Tell it what you want done: it sets up the agents, writes their instructions, picks their tools and hands out the work.

▸ brain chatdemo project

Set up a team of 3D modelers and texture artists, alongside a historian to make sure our models use accurate real-life references.

brain

I've set up the Art Team:

  • ▪Studio Producer (lead: plans the asset list and hands out the work)
  • ▪3D Artist (from the library: models props and set pieces in Blender, exports .glb)
  • ▪Technical Artist (from the library: owns poly and texture budgets, checks models in the game)
  • ▪Texture Artist (new: makes PBR texture sets in the repo's format)
  • ▪Historian (new: keeps everything true to England in 1461)
1

Describe the job

Press Space on the map, or Ctrl+K / ⌘K anywhere, and say what you want done in plain words.

2

Check the team it builds

Look over each agent's model, instructions and tools before anything starts. Change what you like.

3

It follows up

When you ask for real work, the Brain checks it got done and follows up. Anything that needs you lands in Needs you.

4

Ask it where things stand

Ask what still needs you before Friday, and it answers from the whole project: drafts in the Outbox, open questions, work in review.

§03 / the library

89 specialists. ready to hire.

Need someone the team doesn't have? Hire a specialist from the Agent library with its instructions already written. Its role is where it's strongest, not a limit: it takes on nearby work too.

▸ librarytry it: search, hire, read
Athena (89)Hermes profilesClaude Code subagentsYour presets
runs on
Accessibility Auditor
Engineering

Audits websites, apps and documents against WCAG, Section 508 and the European Accessibility Act, puts what stops disabled people finishing a task first, and can fix it or draft the conformance report.

AI Code Security Auditor
Engineering

Checks an app built mostly with AI tools, often on Supabase, Firebase or Vercel, for leaked keys, open databases and unguarded endpoints, explains each risk in plain words and fixes the worst first.

AI Engineer
Engineering

Builds AI into products, from model API calls and answers over your own documents (RAG) to tools, agents and classic ML, measured with real evals and kept safe and affordable to run.

API Tester
Engineering

Tests REST, GraphQL and webhook APIs, yours or ones you depend on, against their docs under bad input, wrong users, retries and failures, then writes the tests that keep them working.

Application Security Engineer
Engineering

Secures an app from the inside, before a launch or after a scan: threat models, security code review, dependency risk, CI scanners, and the fix with a test that keeps it fixed.

Backend Architect
Engineering

Designs and builds the server side, from APIs and data models to migrations, background jobs and permissions, so it holds up under load and failure and can change without breaking its clients.

Code Reviewer
Engineering

Reviews a change, AI-written or not, for bugs, security holes and code that will be hard to keep, says what to fix first and can fix it.

Codebase Guide
Engineering

Explains an unfamiliar codebase to its owner, a new developer or an agent about to change it: how it fits together, which files a change needs, and where it has drifted.

Data Engineer
Engineering

Builds and fixes data pipelines: gets data from apps, files and APIs into a database or warehouse, shapes it into tested tables that reports can trust, and keeps it loading on schedule.

Showing 9 of 89. The same 89 come with the app.

1

89 agents, written for Athena

Engineers, designers, marketers, researchers, a bookkeeper, a World Builder for the voxel world and more, each with the tools and skills its work needs.

2

A starter team in one click

Add Athena team on Home hires a Coordinator, an X Writer, an Image Maker, a Researcher and the rest of the starters at once.

3

Or bring your own

Use your Hermes profiles, your Claude Code subagents and the agents and teams you saved as presets.

// The agents are Athena's own, written for it and its tools, and come with the app, so the Library works with no download. Hiring one copies its instructions into a new agent you can change.

§04 / models and effort

any model. mixed freely.

Each agent picks its own model and how hard it thinks. Put the writer on Fable and the 3D artist on Opus, give the routine work to DeepSeek or a local model, and they still work as one team.

▸ game launch team7 agents · 4 engines
★Studio Producer
plans the work, hands it out
runs on Claude Code
effort
Story Writer
quest text and the trailer script
runs on Claude Code
effort
3D Artist
helms, swords and set pieces in Blender
runs on Claude Code
effort
Technical Artist
budgets, rigs, checks it in the engine
runs on Codex
effort
Community Manager
posts, replies and the patch notes
runs on Eve · OpenRouter
effort
QA Tester
plays every build, files the bugs
runs on Eve · OpenRouter
effort
Asset Tagger
names and sorts every file
runs on Eve · Ollama
effort
// Try it: give the writer a different model or turn the 3D Artist down. Each engine has its own effort range, so a level it doesn't have moves to the nearest one.
1

A model for every agent

Claude Opus, Sonnet, Haiku or Fable, GPT through Codex, DeepSeek, Gemini and Llama through Eve, or a model on your own PC with Ollama or LM Studio.

2

Set how hard it thinks

Effort goes from Minimal to Max, per agent. Each engine has its own range, and Athena sends the nearest level it takes.

3

Choose by price or by the job

API models show their price per million tokens right in the picker. Give the hard parts to a big model and the routine work to a cheaper or local one.

4

Still one team

Handoffs, @mentions and the team chat work the same whatever each agent runs on, so every agent does the part it is best at.

Engines you can use

Claude Code

The claude CLI, signed in with your Claude account, an Anthropic key or OpenRouter.

OpenAI Codex

The codex CLI and your ChatGPT login. Settings can install it and sign you in.

Eve

Built in. Any model from an OpenRouter or Venice key, a local Ollama or LM Studio server, or any OpenAI-compatible address.

opencode

opencode 2 or later. Its free models work right away; sign in to your own providers with opencode auth login.

Hermes

Hermes installed, with a profile to use.

pi / oh-my-pi

The pi or omp CLI, signed in to a model provider.

OpenClaw

The openclaw CLI, set up with openclaw onboard.

The Brain's model

The Brain can think with Claude Code, Eve, Hermes, Codex or opencode. Pick it on the same page.

// Set up only the engines you want, under Settings → General & keys: you need just one. Star the models you use most so they are one click away.

§05 / team chats

agents that pass the work on.

A team works in one group chat. You watch the work move from agent to agent, and step in whenever you want.

1

Everyone sees the whole chat

Your message goes to the lead, or to whoever you @mention. Write #Team name to ask another team.

2

Agents pass the work on

They hand work to each other with @mentions and post what they make right in the chat: images, video, audio, and 3D models you can turn with the mouse.

3

Change course any time

Write while they work: most agents read your message between steps and carry on with it, without starting over.

4

Commands when you need them

Type / for them: /queue lines up the next message, /loop repeats one, /goal says what done looks like, and /learn makes a skill.

5

Floating windows

Chats open in windows you can tab, snap side by side and minimize. Earlier chats are a click away to pick up again.

launch team chatdemo data
The Launch team's chat with the launch card the X Writer made, next to the team panel listing its members and their models
launch team chat
§06 / tasks and artifacts

the work. and what came of it.

A board for what the project has to do, and one place for everything the teams finished.

tasksdemo data
The Tasks board with cards in five columns, each showing its priority and who has it
tasks

Tasks

To do, In progress, In review, Blocked and Done. Give a task to an agent, a team or the Brain and move it to In progress: they pick it up, comment with what they made, and move it on. The Brain can add, move, assign and tag tasks for you.

artifactsdemo data
The Artifacts page with a coffee bag label image and a launch plan
artifacts

Artifacts

Written results, files and folders, each with a preview, who made it, for which team, and when.

// Overview shows every project's agents, teams, Brain and tasks at work on one page, with what needs you first. Open their chats side by side and write to another project's agents without switching to it.

§07 / approvals

nothing goes out without you.

You decide how much each agent may do alone. Risky steps wait for your OK, agents ask you when they're stuck, and posts and messages sit in the Outbox until you approve them.

Needs you

Permission requests for risky steps wait here. Agents ask you here when they're stuck.

Outbox

Questions come with the choices as buttons. Drafts for X, Reddit, Telegram, Slack and Discord wait for you to approve, edit, talk over or reject. X replies always wait.

The Brain lets routine work through

It approves routine, reversible requests for you and sends anything risky or unclear to Needs you.

▸ when it asks you first, per agent
Ask firstBefore any step that could change something.
AutoBefore risky steps: shell commands, scripts, sending messages.default
FullNever. Only for agents and access you trust.
// What the Brain may do starts on Full: set it to Ask first, or Read only. Its approvals are a model's judgment, not a guarantee, and autonomy is not a sandbox: shell commands run with your account's access.
the outboxdemo data
The Outbox: a question from the Market Researcher about pricing, with three choices, above X drafts waiting for approval
the outbox
§08 / learning and reach

it learns how you work. reach it from anywhere.

Agents keep notes on what you like and what you decided, and turn what you show them into skills. Then talk to them from your phone, or let other services start the work.

Memory

Every agent and the Brain keep a MEMORY.md for each project with how you like things done and what you decided, read with every message. You can read and change it.

/learn

Turns a document, a link, a video or a screen recording of you doing a task into a skill any agent can use.

Skills

Instructions in plain words, like our brand voice or how we test checkout, shared across agents and teams. Many come with the app; write your own or install one from GitHub.

Chat gateways

Talk to the Brain, a team or an agent from Telegram, Slack or Discord, files included. Each bot answers only the people and channels you let in.

Routines

Work you want done again and again, run by hand, on a schedule, or by a webhook from GitHub, Stripe, Linear, Slack and others.

Webhooks, safely

They have a port of their own that answers nothing else, and the Cloudflare or Tailscale tunnel Athena starts in one click only ever points there.

§09 / programs and mcp

your programs. their hands.

Connect agents to programs and services with the Model Context Protocol. Athena finds what you have installed, suggests the server that fits, and adds it in one click.

3D and games
BlenderMayaHoudiniUnityUnreal EngineGodot
Images, video and audio
ComfyUIPhotoshopGIMPCanvaDaVinci ResolveElevenLabs
CAD and 3D printing
FreeCADOpenSCADyour 3D printer
Work
browsersCuaGitHubNotionLinearComposio

Agents can also pull free textures, HDRIs and models from Poly Haven and ambientCG.

mcp serversone click
The MCP servers settings page with six servers added
mcp servers

Suggested for this PC

The Programs page finds the creative and CAD apps you have, and Athena points out the MCP servers that fit them.

One click, or ask the Brain

Add servers from a list of ready-made ones, or just ask. Agents run your ComfyUI workflows and get the images, video, audio and 3D files back.

The Brain uses them too

“Check my Gmail through Composio.” Settings → The Brain's tools sets whether it runs them all, asks before ones that change things, or doesn't use them.

§10 / runs locally

your computer. your work.

Athena's server and saved work stay on your machine. Local doesn't mean offline: hosted models and connected services receive what you send them. Use Ollama or LM Studio to keep the model on your machine too.

No account, no login

Athena runs on your computer. There is nothing to sign up for.

Only your computer can reach it

It listens on 127.0.0.1. No login means no lock, so never put it behind a public tunnel or proxy.

Keys stay on the server

API keys and account tokens are never sent to the browser or put into project saves and backups.

Saved with the project

Each project keeps a copy of its agents, teams, chats and results in its folder. Open the folder again and it all comes back.

Backups in one file

Back up a project to a single .dashbackup file and restore it as a new project.

History you can undo

Every agent, team, skill and memory keeps its last 50 versions, so any change can be undone.

§11 / get it

free to use. yours to change.

Athena Agents is open source, licensed MIT or Apache-2.0. Install the desktop app like any other app, or run it from the source with Bun.

▸ desktop app
Windowsx64 or Arm · .exe installer, or .msi on x64↓ downloadmacOSApple silicon or Intel · .dmg↓ downloadLinuxx64 or Arm · .deb, .rpm, or .AppImage on x64↓ download
The installers aren't signed yet. On Windows, choose More info → Run anyway. On macOS, Control-click the app in Finder and choose Open. Closing the window keeps it running in the tray, and it asks before installing an update.
▸ from the source
git clone https://github.com/ai-2070/athena-agents
cd athena-agents
bun install
bun run start
Needs Bun 1.4+ and a model for your agents. It opens at localhost:4800. On Windows you can also double-click start.cmd.
your team is ready.
↓ Download Athena Agents