Agent App Marketplace
Apps are how agents get work done.
Apteva apps install in one click. Each one gives agents new tools to call, channels to speak through, UI to surface, memory to keep, workers to run, or sensors to read and motors to drive. From a CRM in the cloud to a robot on the factory floor.
Core apps
The foundations.
The apps agents reach for first. Storage for files, jobs for scheduling, code for repos, CRM for contacts. Other apps build on top — image-studio writes to storage, media catalogs storage, jobs schedules across them all.
Storage
OfficialcoreFile storage with signed URLs and three visibility tiers — private, signed, public. Range requests, ETag, Cache-Control. Local-disk today; S3 / R2 / MinIO behind the same key abstraction next.
Jobs
OfficialcoreScheduled-job runner. Apps and agents enqueue work for later — once, on an interval, or on a cron expression. At-least-once delivery, idempotency keys, exponential backoff, runs log.
Code
OfficialcoreRepositories — named, framework-tagged code workspaces. First-class editing tools modelled on Claude Code: partial reads with line numbers, exact-match string edits, atomic multi-edit, repo-wide grep + glob.
CRM
OfficialcoreContacts store with multi-value channels, typed custom attributes with provenance, append-only activity log, soft-delete + merge. The agent reads and writes the same data your team does.
More first-party apps
Productivity, channels, observability, and a few specialised utilities — all maintained by Apteva.
Tasks
productivityMission board. Agents create and complete tasks via MCP; the dashboard tracks progress live.
Status
observabilityPer-instance status line. Agent writes 'what I'm on right now'; dashboard reads it live.
Channel Chat
channelsChat bridge between agent and dashboard or external channels. Persisted messages with monotonic ids for clean reconnect.
Simple
uiMinimal read-only UI for watching an Apteva agent — embed anywhere on the same host. Path-mounted SPA, no sidecar.
Media
mediaCatalog + cheap derivations for audio/video/image files in Storage. Probes new files with ffprobe, generates thumbnails and waveforms. Built on Storage.
Image Studio
creativeGenerate images via any compatible provider — OpenAI today, more as the catalog grows. Optionally saves outputs to Storage for permanent, shareable references.
Trading
financePaper trading desk. Multi-portfolio, multi-asset (equity, crypto, polymarket). Deterministic paper-execution engine, daily-loss halt — no broker, no real money.
Email Checker
utilityStateless email validation. RFC 5322 syntax, DNS MX lookup, optional SMTP RCPT probe, plus list-based classification — disposable, free, role.
One-click install
Three steps to a new capability.
Run apteva
npx apteva spawns the server and dashboard locally. Or use Cloud Hosting and skip the install.
Click install
Browse the marketplace, click Install on an app. Dependencies cascade — anything it needs comes along.
Agents use it
New tools appear in the agent workspace automatically. UI panels show up in the dashboard. Channels and workers start routing.
npx aptevaOpen source. Self-hosted. Apps install through the dashboard.
What an app can be
Six surfaces. Mix any combination.
An Apteva app is a manifest plus whatever surfaces it wants to provide. Some apps are pure tool packs. Others are full sidecars with their own UI, channels, and workers. These are the surfaces for server-installed apps — hardware-native tools register directly through the embeddable core (covered below).
MCP Tools
Tools the agent calls. Each tool is a function the app exposes — query a CRM, post a task, send a message.
HTTP Routes
Reverse-proxied at /apps/<name>/* — the app serves its own API alongside the platform.
UI Panels
Bundled into the dashboard slot. The operator watches and configures the app from one place.
Channels
Slack, email, Telegram, custom — inbound messages wake the right thread, outbound replies route through.
Workers
Background goroutines on a cron-style schedule. Things the agent shouldn't have to remember to do.
Prompt Fragments
Concatenated into the agent's directive at boot. The app teaches the agent how to use it.
Across every domain
Apps for every kind of work.
Some apps are full UIs you watch. Some are MCP tool packs the agent calls. Some are channel adapters, hardware drivers, or background workers. Across the marketplace they cover business operations, personal life, robotics, industrial control, financial markets, and creative work — anywhere persistent agents have a job to do.
Business
Run the operation- CRMUI + tools + memory
- TasksUI + tools
- Helpdesktools + channel
- Invoicingtools + workers
- AnalyticsUI + tools
Personal
Run your own life- Inboxchannel + tools + UI
- Calendartools + workers
- Habitstools + UI
- Journalmemory + tools
- Moneytools + workers
Robotics
Embodied agents- Motor Controlembedded tools
- Lidarembedded tools
- Visionembedded tools
- Navigationembedded tools
- Manipulationembedded tools
Industry
Plant + factory floor- PLC Bridgetools + workers
- SCADAUI + tools
- Maintenanceworkers + channel
- Qualitytools + memory
- Energytools + workers
Trading
Markets, around the clock- Market Datatools + workers
- Brokertools
- Risktools + UI
- BacktestUI + tools
- Signalstools + channel
Creative
Make and ship- Image Gentools
- Audio Studiotools + UI
- Video EditorUI + tools
- Asset Librarytools + memory
- Publishingchannel + workers
Examples of what fits the marketplace. Apteva ships the platform and the first-party set above; everything else comes from the ecosystem and from anyone who builds with the SDK.
Stack apps. Run an operation.
A few apps, an autonomous team.
Apps compose. Install a CRM, a task board, a Slack channel, a billing integration — the agent uses all of them, coordinated, around the clock.
Support team
Agent triages tickets, drafts replies, escalates, follows up — across email, chat, and your knowledge base.
Sales pipeline
Lead enrichment, personalized outreach, follow-ups the agent schedules itself, deal updates pushed to your team.
Content operations
Research, draft, publish, distribute, measure — an editorial team that runs around the clock.
Autonomous robot
Embedded core, native hardware tools. The agent paces itself between sensor reads and motor commands — same loop, physical work.
Beyond the server
Same agent runtime. Cloud, edge, or robot.
On a server, apps install through the marketplace as sidecars and static UIs. On a robot, IoT gateway, or industrial controller, the core compiles in directly — and the "apps" are native Go tools the agent calls. Same continuous loop, same workers, same self-pacing. The deployment changes; the agent model doesn't.
Server
Marketplace apps
Sidecars + static UIs installed through the dashboard. Cascade dependencies. Reverse-proxied at /apps/<name>.
Edge / Gateway
Embedded core + local tools
Single Go binary, no external runtime. Local thinking with cloud sync when reachable. Survives flaky uplinks.
Robot / Device
Embedded core + hardware drivers
Native motor, sensor, camera tools registered directly with the engine. Sub-second observe → reason → act loop.
See for the hardware path.
Build your own
An SDK, a manifest, a sidecar.
Apps depend on the public Go SDK only — never on apteva-server internals. Declare what you provide in apteva.yaml, implement the surfaces you care about, ship a binary or a static bundle. The platform handles the rest.
Apps are Go binaries. They start in milliseconds, eat single-digit MB of RAM, and run at native speed. Stack a dozen apps on the same server — the footprint stays small.
package main
import sdk "github.com/apteva/app-sdk"
type App struct{}
func (a *App) Manifest() sdk.Manifest { return loadManifestFromYAML() }
func (a *App) MCPTools() []sdk.Tool {
return []sdk.Tool{{
Name: "hello",
Handler: func(ctx *sdk.AppCtx, args map[string]any) (any, error) {
return "hi from " + ctx.Manifest().Name, nil
},
}}
}
func main() { sdk.Run(&App{}) }Install your first app.
npx aptevaMarketplace ships in the dashboard. Browse, click install, the agent picks it up.