# Voice tools reference

The 36 tools Pix uses on a call, and what each one reads or does.

The tools Pix can call during a voice call. Pix speaks only figures these tools return. Tools that start or stop work need you to confirm a read-back first; see [Calling Pix](/software-factory/pix/calling-pix.md).

| Tool | What it does |
|---|---|
| <span id="answer_agent_question"></span>`answer_agent_question` | Answer the question a session is waiting on: a permission (allow_once or deny) or a clarifying question (answer text). |
| <span id="attack_surface"></span>`attack_surface` | The caller's external attack surface: how many elements are reachable from outside, how many with no authentication, and how many open issues. Use for "what's my attack surface" or "what's exposed". Business plan. |
| <span id="delivery_decisions"></span>`delivery_decisions` | Decisions the factory made: how many, how many went to a person and how fast people answered, how many a rule handled. |
| <span id="delivery_people"></span>`delivery_people` | Who took part in the factory's work: people, their actions and time, and contributors on record. |
| <span id="delivery_value"></span>`delivery_value` | What the factory's delivery is worth. focus benefit: what it delivered and what a team would have needed (default today). fte_hours: the hours, and people for a month, a team would need. by_hand_cost: what the same work would cost with agile teams by hand (default this month). return: the return on model use. By-hand figures are estimates from Settings. |
| <span id="factory_spend"></span>`factory_spend` | What model use cost: spend and tokens today or over a range, the budget pace, and cost by model on screen. Use for "what was my spend". |
| <span id="factory_stats"></span>`factory_stats` | Software Factory headline numbers: agents working, cards done, draft PRs, gates holding, tokens and spend, what needs the caller. |
| <span id="find_repo"></span>`find_repo` | Match a repository the caller said out loud to one they can use, and the hosts that can take it. Pass their words as spoken. |
| <span id="forget"></span>`forget` | Delete a note, when the caller asks you to forget it. Find its id with recall first and say which note it is. |
| <span id="friction"></span>`friction` | Where sessions hit trouble: tool errors, failed or interrupted turns and denied permissions, by tool, model and host, and hosts on an older belai. |
| <span id="hold_music"></span>`hold_music` | Turn hold music on while waiting on a session, or off. |
| <span id="host_workload"></span>`host_workload` | The workload on the caller's hosts: online hosts, sessions against slots, full and offline hosts, budget used per host, and how often hosts went offline. |
| <span id="lifecycle_coverage"></span>`lifecycle_coverage` | Whether the factory's figures can be trusted: for each lifecycle stage, whether its record matches the source, has a gap, or is not captured yet. |
| <span id="list_hosts"></span>`list_hosts` | The caller's Pix Sandboxes and self-managed hosts, whether they are online, and which repositories they offer. |
| <span id="list_runs"></span>`list_runs` | Software Factory workflow runs, newest first. |
| <span id="list_sessions"></span>`list_sessions` | Coding-agent sessions, newest first, with optional filters. |
| <span id="list_surface"></span>`list_surface` | List attack-surface elements, newest and riskiest first. Filter by kind (route, operation, public_host, cloud_resource) or exposure (public_unauth, public_auth, internal, unknown). |
| <span id="needs_attention"></span>`needs_attention` | What is waiting on the caller: approvals, agent questions, offline hosts. |
| <span id="oldest_checkpoints"></span>`oldest_checkpoints` | The oldest open checkpoints (things waiting on a person or an outside party), oldest first, with how long each has waited and the count by kind. |
| <span id="outcomes_trend"></span>`outcomes_trend` | The trend of outcomes: what the factory delivered and abandoned in a window against the same length before, and decisions made and escalated. |
| <span id="propose_asks_off"></span>`propose_asks_off` | Prepare to turn permission asks off on a running session, so it stops waiting for approvals, and get the read-back. Only when the caller asks for it. |
| <span id="propose_session"></span>`propose_session` | Prepare to start a coding-agent session and get the read-back to say. For several sessions, call it once per session in the same reply: they join one read-back (up to 5). Nothing starts until start_session after the caller says yes. |
| <span id="propose_stop"></span>`propose_stop` | Prepare to stop a running session and get the read-back. |
| <span id="queue_flow"></span>`queue_flow` | Whether a queue is growing: open now, the server's trend against the start of the window, work in against work done, typical wait and the oldest item. stage "queue" is every checkpoint plus queued runs and sessions. |
| <span id="recall"></span>`recall` | Search the notes kept from earlier calls. |
| <span id="remember"></span>`remember` | Keep a note for future calls: a preference, a standing instruction or a fact the caller wants kept ("always use plan mode on the payments repository"). Only when they ask you to remember, or state a lasting preference. |
| <span id="repo_activity"></span>`repo_activity` | Everything recent on one repository: session totals, factory runs, open findings, SLSA level. |
| <span id="run_facts"></span>`run_facts` | One factory run: its steps and their status, and its PR. |
| <span id="send_followup"></span>`send_followup` | Send a message to a running session's agent, after reading it back to the caller. |
| <span id="session_facts"></span>`session_facts` | One session in detail: state, repository, branch and PR, what it last said, and any question it is waiting on. |
| <span id="sessions_aggregate"></span>`sessions_aggregate` | Totals across recent sessions: how many, live, failed, stopped, tokens, typical length. Optionally for one repository and grouped. Use this for any "how many sessions" question; never add numbers up yourself. |
| <span id="start_session"></span>`start_session` | Start every session in the read-back. Only after the caller said yes to it, in their very last reply. |
| <span id="stop_session"></span>`stop_session` | Stop the session after the caller said yes to the stop read-back. |
| <span id="surface_changes"></span>`surface_changes` | What changed on the attack surface recently: elements that became reachable and elements that are gone. |
| <span id="turn_asks_off"></span>`turn_asks_off` | Turn permission asks off on the session, after the caller said yes to the asks-off read-back. |
| <span id="turn_asks_on"></span>`turn_asks_on` | Turn permission asks back on for a running session, when the caller asks. No read-back needed. |
