hoziron-cli memory

Manage agent memory — the per-agent key-value store.

Synopsis

hoziron-cli memory <subcommand> [options]

Subcommands

SubcommandDescription
list <agent>List KV pairs for an agent
get <agent> <key>Get a specific value
set <agent> <key> <value>Set a value
delete <agent> <key>Delete a KV pair

hoziron-cli memory list

hoziron-cli memory list <agent> [--json]

Example

$ hoziron-cli memory list claims-agent
KEY                        VALUE
claims_processed_count     142
avg_processing_time_ms     3200
last_claim_number          CLM-2026-142
escalation_count           7
shift_start                2026-06-04T08:00:00Z

$ hoziron-cli memory list claims-agent --json
{
  "claims_processed_count": 142,
  "avg_processing_time_ms": 3200,
  "last_claim_number": "CLM-2026-142",
  "escalation_count": 7,
  "shift_start": "2026-06-04T08:00:00Z"
}

hoziron-cli memory get

hoziron-cli memory get <agent> <key> [--json]

Examples

$ hoziron-cli memory get claims-agent claims_processed_count
142

$ hoziron-cli memory get claims-agent last_claim_number
CLM-2026-142

$ hoziron-cli memory get claims-agent nonexistent_key
(not set)

hoziron-cli memory set

hoziron-cli memory set <agent> <key> <value>

Examples

$ hoziron-cli memory set claims-agent escalation_threshold 5
✓ Set claims-agent.escalation_threshold = 5

$ hoziron-cli memory set claims-agent notes "Agent paused for maintenance at 10:00"
✓ Set claims-agent.notes = "Agent paused for maintenance at 10:00"

hoziron-cli memory delete

$ hoziron-cli memory delete claims-agent notes
✓ Deleted claims-agent.notes

Memory Isolation

Each agent has its own isolated memory scope:

  • Agent A cannot read or write Agent B's memory
  • This is enforced by the kernel's BoundMemoryHandle, which is structurally bound to one agent's ID at construction — no method accepts a free agent_id parameter, so cross-agent access is impossible through it rather than merely rejected at runtime
  • In workflows, data passes between agents only through step outputs (with PII tokenization)

How Agents Use Memory

Agents read/write memory through their equipped skills and competency logic. Competencies define [[metrics]] that reference memory keys:

[[metrics]]
label = "Claims Processed"
memory_key = "claims_processed_count"
format = "number"

The agent increments this key as it processes claims. The dashboard and memory list command expose the current values.

Use Cases

PatternExample
Countersclaims_processed_count, errors_today
Last-seen statelast_claim_number, last_run_at
Configurationescalation_threshold, auto_approve_under
Operator notesnotes, maintenance_reason
Agent preferencespreferred_format, verbose_mode

See Also