Home / Features / Agents and automations

Build AI agents and automations with osFoundry

Persistent agents that schedule, plan, use tools, share memory, and run 24/7 — without you watching.

osFoundry is an agent platform with persistent sessions, scheduled automations, scoped tool access, shared memory groups, and a plan mode for hard tasks. Agents you build run on a cron, monitor inboxes, watch for changes, and ping you only when something needs your eyes. Profile-based roles let one workspace host a code-reviewer agent, a customer-success agent, and a research agent — each with its own scoped tools and memory.

Quick answer

Persistent agent sessions — agents keep state between runs, not just within one chat.
Schedule any agent to run on a cron — every minute, hour, or day.
Per-agent tool scope — restrict which tools (web, email, code, shell) each agent can call.
Shared memory groups — agents read/write to the same knowledge base.
Plan mode — for complex tasks, the agent proposes a plan first and you approve it.

Key capabilities

Profile-based agents (system prompt, tools, memory groups, model) — versioned in osStudio.
Automation editor — define triggers (cron, webhook, channel event) and the agent that runs.
Tool catalog — web search, browser, email, file ops, shell, code, custom MCP tools.
Multi-agent orchestration — agents call other agents as sub-tasks.
Live log streaming and replay for every agent run.
Pause / resume / cancel for in-flight runs.

How to do it in osFoundry

Create an agent profile — In osStudio → Agents, define the system prompt, pick the tools the agent can use, attach memory groups, and pin a model.
Test interactively — Open a session in /agents and chat with it. Watch tool calls live, edit the system prompt, iterate.
Schedule it — In /agents/automations, wrap the profile with a cron trigger and a starting prompt. The agent runs unattended.
Get notified — Add an explicit "ping me" step (Slack message, email, in-app notification) at the end of the agent’s task.

Frequently asked questions

Are osFoundry agents truly persistent?

Yes. Sessions persist between runs in the agents store and the workspace database — agents resume context, not start fresh.

How are scheduled agents billed?

You pay only for the seconds an agent actually runs (compute) and any tokens it sends (BYOK to your provider). Idle agents cost nothing.

Can one agent call another?

Yes — multi-agent orchestration is built in. The calling agent treats the other agent as a tool with a typed input/output.

How do I keep an agent from doing something it shouldn’t?

Tool scope per agent (only allow the tools it needs), workspace policies (block specific actions), and plan mode (require your approval before destructive ops).

Can agents use my knowledge bases?

Yes — attach memory groups to the agent profile. The agent gets RAG-style read access to those KBs and can write back too if you allow it.