Agent platform

Build and operate your own agents — no-code definitions, triggers, approvals, connectors, secrets, and a full audit trail, all from the Agents surface.

The Agent platform — Agents in the navigation, headed Agent Control Center — is where you go beyond the built-in roles and run your own agents: define what an agent knows and may do, schedule it, review what it did, and gate anything risky behind human approval.

Agent modes

Every agent runs in one of three modes:

Mode	Behavior
Interactive	You drive the conversation; the agent acts turn by turn
Assisted	The agent works through a task but pauses for your input at key points
Autonomous	The agent runs unattended on triggers

Autonomous mode requires a Pro or higher plan. See Billing.

The tabs

Agents

Define agents with no code:

• Persona — written in Markdown: who the agent is, what it’s for, how it should behave.
• Allowed tools — exactly which tools this agent may call.
• Model — which AI model powers it.
• Permission mode — ask (request approval for side effects), auto (act within its allowed tools), or read-only (never change anything).

Runs

Every execution appears here with live status: queued → running → awaiting approval → completed / failed. Open a run to see the full transcript and tool activity. Runs have shareable links, so you can send a colleague exactly what an agent did.

Approvals

Side-effecting actions — Salesforce writes, deployments, outbound calls — don’t execute immediately. They park as pending approvals. You review the exact input, then approve or deny.

When you approve an action, the stored input executes exactly as you approved it — the agent can’t swap the payload after the fact.

Triggers

Start runs three ways:

Trigger	Notes
Cron schedule	Recurring runs, with 15-minute resolution
Webhook	External systems start a run by calling an endpoint
Manual	Run on demand from the UI

Skills

Reusable know-how that agents load — procedures, conventions, domain knowledge. Write a skill once, attach it to many agents.

Marketplace

Publish agents and skills for other organizations, or install ones shared by others — a way to distribute proven automations across orgs.

Connectors

Give agents reach beyond LivingContext by registering outbound endpoints:

• MCP servers — connect external MCP tools
• HTTP — plain HTTP endpoints
• OpenAPI — endpoints described by an OpenAPI spec

Secrets

Store encrypted values (API keys, tokens) that connectors and tools need. Secrets are injected server-side at call time and never shown to the model — an agent can use a credential without ever seeing it.

Channels

Reach the agent where your team already works: connect Slack so agents can respond in your channels (see Slack integration), post a card to Microsoft Teams when a background run finishes via an incoming webhook, and mint public API keys for driving runs from your own tools.

Memory

Long-term memory agents accumulate across runs — decisions, facts, workflows, and blockers. You can inspect and curate what an agent remembers.

Audit

An immutable log of every side-effecting action any agent has taken: what ran, who approved it, what it touched. This is your compliance trail.

Billing

Usage and credits for agent runs. See Billing.

Programmatic access

Create public API bearer keys to start runs from your own systems — CI pipelines, internal tools, or scheduled jobs outside LivingContext. Combined with webhook triggers and shareable run links, you can wire agents into existing workflows end to end.

Governance in one picture

1. An agent’s allowed tools and permission mode bound what it can attempt.
2. Approvals gate every side effect in ask mode.
3. Secrets keep credentials out of model context entirely.
4. Audit records everything that actually happened.

Start new agents in read-only or ask mode. Promote to auto only after the Runs history shows consistent, correct behavior.