Agents

Agents that remember, learn, and stay within boundaries

DeckCrew agents combine persistent personality, shared memory, trust-aware access, and proactive routines so they can help with real work instead of behaving like generic chat sessions.

Soul

Every agent has a persistent way of working

Agents are not just prompts with a name. DeckCrew gives each one a persistent behavioral profile so support, sales, development, and content work can feel intentionally different.

  • Stable behavior Keep a consistent style and decision pattern over time.
  • Role fit Tune an agent for careful support, structured research, or repo-safe engineering.
  • Context overlays Adjust behavior for different channels, customers, or teams.
  • Revision history Inspect changes and roll back if the agent starts drifting.

Persistent soul with scoped overlays and revision history. Agents can introspect, compare changes, and restore previous soul versions when behavior needs to be corrected.

Memory

Agents build on more than the current chat

DeckCrew agents can use the current conversation, customer history, shared company knowledge, and proven runbooks so they do not start from zero every time.

  • Current conversation Use what just happened in the active task.
  • Customer continuity Carry over context from earlier interactions when it should matter.
  • Shared logbook Use the same company knowledge across multiple roles.
  • Proven patterns Lean on runbooks from work that already succeeded.

Session context, local continuity, shared logbook retrieval, and runbook-based pattern recall combine into the runtime context pack with visible attribution.

Trust

Access is governed outside the prompt

Agents stay safer because their boundaries do not depend on the prompt behaving well. Tool access, approval requirements, and trust checks are enforced by the platform.

  • Per-agent boundaries Each agent only gets the tools and actions it should use.
  • Approval-aware actions Draft first and require review before risky mutations.
  • Trust by identity External and internal users can have different privileges.
  • Traceable decisions Review what the agent used and why it took action.

Platform-enforced allowlists, trust-required capabilities, approval flows, and execution telemetry keep risky behavior outside the agent's control.

Presence

Agents can keep watch between conversations

DeckCrew agents can check for changes, failures, or follow-up work on a schedule and report what matters without constant prompting.

  • Scheduled routines Run hourly, daily, or weekly checks.
  • Useful summaries Report what changed instead of sending noise.
  • Quiet windows Stay out of the way outside active hours.
  • Escalation when needed Surface problems with the right context already attached.

Engine-managed proactive presence supports pulse, digest, and attention flows with active-hours logic and lookback context.

Agents FAQ

Questions about DeckCrew agents

What makes a DeckCrew agent different from a normal AI assistant?

DeckCrew agents keep shared company knowledge, role boundaries, approvals, and memory over time. They are designed for real work, not just one-off prompting.

How does memory work across conversations?

Agents remember across the current session, customer history, shared team knowledge, and proven runbooks. Replies can show where that context came from.

Can agents work without being prompted?

Yes. Agents can run on schedules, check for issues, and prepare updates or summaries without waiting for someone to ask first.

How do approvals and trust fit in?

Sensitive actions can stay behind approval, and trust can be checked per user and per tool. That makes agents safer to use across teams and customer-facing workflows.

Where do the role-specific pages live?

Support, sales, development, website, and content pages live under the use-cases section. The agents page explains the core model behind them.

Ready to set up your first agent?

  • Guided onboarding included
  • Your own provider keys
  • No commitment required