product update

CAI v1.0.5: clearer guidance, smoother operation, and early support for long-running workflows

CAI v1.0.5 brings clearer in-product help, a smoother terminal experience, and early support for long-running cybersecurity workflows.

Alias Robotics

• 3 min read
Cybersecurity AI CLI interface update showing improvements in guidance, reliability and long-running workflows in CAI v1.0.5

Today we’re releasing CAI v1.0.5, a focused update that improves something fundamental in day-to-day use:

what the documentation says, what the CLI exposes, and how CAI actually behaves are now aligned.

This reduces guesswork, removes inconsistencies across references and makes CAI more predictable in real-world environments.

It also introduces early support for long-running operations—still maturing, but pointing toward sustained usage over time.

Command reference and in-product help aligned with real behavior

One of the core improvements in v1.0.5 is a more coherent documentation layer across CAI.

Instead of fragmented or outdated references, operators now work with:

  • A canonical command catalog that reflects what the CLI actually registers
  • Consistent in-product help for slash commands (/help, /h/<command>)
  • Documentation that matches the real entry points exposed in the REPL

This eliminates common friction points such as:

  • “the docs say X but the REPL behaves differently”
  • outdated command references
  • unclear command hierarchies

Slash commands

Command documentation is now aligned with actual usage:

  • /help and the CLI command reference list only valid, current commands
  • Subcommands are documented where that reflects real behavior (e.g. /model show)
  • Deprecated or misaligned root commands are removed or repositioned

Examples of changes:

  • /model show replaces standalone /model-show
  • /parallel, /queue, /shell and TUI controls reflect actual execution paths
  • Generic or misleading root commands are no longer presented as primary interfaces

Environment variables

Configuration is now clearer and easier to reason about:

  • /env becomes the primary interface (list, get, set, defaults)
  • /help includes full environment reference tables
  • /help var NAME provides detailed explanations for each variable

This ensures that:

  • what you configure matches what the system applies
  • defaults are visible and understandable
  • runtime behavior is easier to predict

Help entry points

Operators now have a consistent navigation model:

  • /help → full guide and environment tables
  • /help topics → compact index
  • /help <topic> → deep dives
  • /help var NAME → detailed variable behavior

What this changes in practice:
Teams onboard faster, scripts and runbooks age more slowly, and misconfiguration becomes less frequent.

A more predictable and reliable CLI experience

v1.0.5 also improves how CAI behaves in real usage, especially during longer sessions.

Key improvements include:

  • More predictable session lifecycle (/exit, end-of-input)
  • Fewer “stuck terminal” situations
  • Startup behavior you can trust (updates are not applied without explicit opt-in)
  • Clearer feedback during long-running operations

The result is a CLI that feels:

  • More stable
  • More transparent
  • More aligned with operator expectations

Safer handling of untrusted content

CAI continues to reinforce a critical principle:

external content should be treated as data, not instructions.

In typical configurations, CAI strengthens this behavior automatically, helping reduce prompt-abuse risks without requiring manual prompt engineering.

This improves security posture without adding friction to workflows.

One entrypoint, flexible control: the Selection Agent

CAI maintains a single conversational entrypoint through the Selection Agent, which routes each task to the most appropriate specialist.

  • Automatic routing reduces friction
  • /agent allows explicit control when needed

What this enables:
Automation where it helps, precision where it matters.

Optional gateway configuration

For environments with specific infrastructure requirements, CAI supports an optional API routing configuration:

  • ALIAS_API_URL allows directing traffic to a custom endpoint
  • Fully documented within the in-product help system

This enables better alignment with:

  • internal network policies
  • custom hosting setups
  • controlled deployment environments

Early support for long-running operations (Beta)

v1.0.5 introduces an initial implementation of periodic operations, designed for sustained oversight over time.

This includes:

  • Guided flows for repeated security work
  • Support for background-friendly execution
  • Optional OS-level supervision

This capability is still evolving and not yet at the same maturity level as standard interactive CAI.

It should be:

  • tested in controlled environments
  • tuned based on intervals and privileges
  • gradually integrated into workflows

Future releases will extend and stabilize this area.

What this enables:
A first step toward CAI-assisted monitoring, with realistic expectations while the feature matures.

Final thoughts

CAI v1.0.5 moves the system toward operational consistency.

It aligns documentation with behavior, improves reliability in real usage, and begins extending CAI into longer-running workflows.

In practice, that means:

  • Less ambiguity in how the system works
  • More confidence during extended sessions
  • Better alignment between operator intent and system behavior

And that’s where CAI continues to focus:
evidence-based, controllable automation for cybersecurity workflows.