Use Cases

Practical starting points for agents that run on a schedule or continue outside one chat window

Comis is most useful when you need to leave an agent working and still keep its state, limits, and failures understandable. The patterns below use capabilities available in the runtime today. They are starting points, not complete industry solutions.

You don’t need to understand the technical details to use this feature. The configuration examples below are copy-paste ready.

Overnight research brief

What you build: a scheduled run gathers sources, compares the findings, asks a critic to check the draft, and delivers a brief in the morning.

Why Comis fits:

• execution graphs coordinate research, comparison, criticism, and retries;
• context recovery lets the run search and expand canonical detail omitted from the active prompt;
• memory history keeps eligible learned claims tied to sources and trust signals;
• configured budgets and cost controls limit admitted work; and
• traces and incident evidence show what completed, what failed, and what it cost.

This is the recommended first workflow because it is useful, runs unattended, and can start with read-only web and file access.

Controlled operational investigation

What you build: an agent collects read-only system evidence, connects related events, and produces an incident report with cost and audit records.

Why Comis fits:

• scheduled checks and heartbeats can trigger bounded investigations;
• scoped tools and capability gates limit which operational surfaces an agent can reach;
• incident explanation connects execution, delivery, policy, and cost evidence; and
• supported approval paths can pause configured privileged actions.

Keep the first deployment read-only. Add remediation only after approval behavior and Linux execution isolation are enabled and tested on the target host.

Routed support and knowledge work

What you build: specialized agents handle different channels, users, or knowledge areas while the operator keeps one set of runtime controls.

Why Comis fits:

• routing bindings choose the agent for each inbound path;
• each agent can have separate context, memory, tools, budgets, secret rules, and workspace files;
• RAG memory retrieves relevant knowledge with trust-aware ranking; and
• delivery and system diagnostics expose failures across channels.

Human escalation and destructive actions still need application-specific workflow design. Comis approval support is not yet a universal policy engine.

When a smaller tool may be enough

If you need a simple chat assistant, a model wrapper, or only a graph library, a smaller tool may be easier. Choose Comis when one or more unattended agents use real credentials, build long context, work on a schedule, or must leave evidence after a failure.

Start building

Quickstart

Install Comis and verify one bounded agent workflow.

Execution Graphs

Coordinate typed sequential and parallel work.

Security Boundaries

Review controls, defaults, and residual risk before adding authority.

Operations

Deploy, monitor, back up, recover, and troubleshoot Comis.