The surface every agent runs with.
Capabilities are what your agents are allowed to touch: inboxes, tools, sandboxes, memory, channels, and streams. The platform pins the release; this is the toolkit inside the run.
Bundled features for every agent.
Agent email
Each agent gets its own inbox. Inbound and outbound mail on the same thread-and-run model as API and channels.
Tool allowlists
The model only selects tools the release declared—one declared tool per step, no ambient access.
Sandboxes
Isolated compute with leases, checkpoints, and controlled egress for untrusted agent work.
Screenshots & OCR
Browser capture and document OCR as first-class tools agents can call when allowed.
Memory
Authorized hybrid recall—not a transcript dump. Grants and lifecycle keep memory useful and scoped.
Channels
Slack, Telegram, and more—plus schedules, heartbeats, and webhooks—on one runtime.
Streams, events, and approvals.
Agents produce durable streams for product UX, semantic events for automation, and clean pause/resume when a tool needs approval—so your app stays in control.
Streaming, web tools, and metering.
Token and event streams your product can render live—sessioned and tenant-scoped.
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Sensitive tools suspend the run for approval without losing continuation state.
Durable streams for UX; semantic events for backends and webhooks.
Secrets and OAuth stay out of manifests and model context—attached at deploy time.
Every tool call and run step is reconstructable; usage meters with the work.
API, email, channels—same threads and runs.
Whatever ingress you use, agents share one execution model. See Platform for how releases and deployments pin that model.
Capabilities describe the toolkit. Platform describes define → publish → deploy → run.
Frequently Asked Questions
Capabilities are what agents can use during a run—email, tools, sandboxes, memory, channels, streams. Platform is how you define, release, deploy, and operate those agents.
The release declares the allowlist; the deployment binds credentials and policies. The model can only select declared tools—one at a time.
Healthcare is a core sector. Deployments.ai is designed for HIPAA-compliant agent deployments—controlled credentials, audit-friendly events, isolated execution, and product-embedded runs you can govern.
Coming soon. Follow @DeploymentsAI on X for launch updates.