- 421 outcomes reached
- 216 vulnerabilities fixed by the agents
- 139,074 decisions settled without a person
- 9 h of people time behind all of it
Vulnetix's own Software Factory, last 30 days. Every figure is on a capture below.
Every part of the factory, one phone call away
This is what a day of running the Software Factory by phone looks like. Each step is something you say, what Pix does with it, and what the factory produces behind the call. The results are our own factory's, from the captures on this page.
- 07:40, on the way in: Find out what happened overnight
- 07:55: Hear what it delivered
- 08:30: Launch work in one sentence
- 09:15: A critical CVE lands. Patch it before lunch
- 11:00, between meetings: An agent needs an answer
- 14:00: Changes that land, and conflicts that do not stop them
- 15:30: Ask what it is costing
- 17:00, end of the week: Ask what it was worth
- Any time: When something looks wrong, stop it
- Afterwards: Every call stays on the record
Find out what happened overnight
For example, you say: "What happened overnight, and what needs me?"
Pix: Pix opens with what is running and what needs you, then answers from the factory's records: which agents are working, what finished, which host went offline, where sessions hit tool errors or denied permissions.
Behind the call: The factory works while you sleep. Schedules start the template workflows (test sweeps, lint sweeps, dependency upkeep, delivery sweeps), and every event, decision and handoff lands on the event stream.
What you get: You walk into standup already knowing what moved and what is stuck.
Hear what it delivered
For example, you say: "How did the week go compared with last week?"
Pix: Pix compares what the factory delivered and abandoned in a window with the same window before, and reads any run's steps, status and pull request. A long list goes to the screen as a table while Pix says the headline.
Behind the call: Every outcome is recorded against the trigger that started it: a pull request merged, an issue closed, a card done. Abandoned work is counted too, so the figure is honest.
What you get: Outcomes, not activity. You hear what shipped, and what did not.
From our own factory: 421 outcomes in 30 days: 20 pull requests merged, 5 issues closed and 315 cards done, from 3,339 sessions and 123 hours of agent time.
Launch work in one sentence
For example, you say: "Sweep all repos and report on every open issue created outside the team."
Pix: Pix matches the repository you named, picks a Pix Sandbox first and then the least busy host, and reads back the repository, branch, host and task. It starts only when your next answer is a plain yes. One read-back can hold up to five sessions.
Behind the call: The session runs on a Pix Sandbox, an isolated machine Vulnetix operates around the clock, or on a host you run Belai on. It records that a Pix call started it, with the task you gave.
What you get: Work that would wait for your laptop starts from the platform.
A critical CVE lands. Patch it before lunch
For example, you say: "Is anyone exploiting the new fast-uri CVE, and which fix do we take?"
Pix: Pix looks the vulnerability up through the read-only tools of the Vulnetix MCP server: exploits, fixed versions, packages and detection rules. Ask it to start a session on the affected repository, and it reads the plan back first.
Behind the call: The security crew takes it: a scout scans and confirms the finding, patchers write the fix, a verifier re-checks the verdict. Every install goes through the Package Firewall and every model call through the AI Firewall. The quality gate and the release jail are plain rules, and the work ends in a draft pull request with its evidence.
What you get: A finding becomes a reviewed fix while you are still away from the desk.
From our own factory: 216 vulnerabilities recorded as fixed by the agents in 30 days.
An agent needs an answer
For example, you say: "Tell the patcher to keep the lockfile and skip the major bump."
Pix: When an agent is waiting on you, Pix says what it asked. Allow a permission once or deny it, or answer in words. Pass a running agent a message and Pix reads it back before sending. Pix never allows an action for good, and plan reviews wait for the screen.
Behind the call: System One scores the routine decisions so they never reach you. What does reach a person lands in Checkpoints, one queue of 11 kinds, oldest first, and the cockpit's Needs you shows its head.
What you get: You answer the few questions that matter, and nothing else interrupts you.
From our own factory: 139,074 decisions in 30 days, all settled without a person. The 17 questions that needed one were answered in Checkpoints.
Changes that land, and conflicts that do not stop them
For example, you say: "Did the patch pass review, and what is its pull request?"
Pix: Pix reads a run's steps and its pull request, and tells you what a session last said.
Behind the call: Reviews are posted on every pull request the factory opens. A pull request that conflicts with its base goes to the conflict resolver, which tries up to 5 sets of choices, tests each and pushes the first that passes. People sign off what is committed in their name, with a key that never leaves AWS KMS.
What you get: Pull requests arrive reviewed and mergeable. A person still merges.
From our own factory: 62 commits in 30 days, 59 signed off by a person, 205 files changed, and 81 reviews posted.
Ask what it is costing
For example, you say: "What have we spent this month, and are we on pace?"
Pix: Pix reads spend and tokens for today, the week or the month, the pace against your budget, and cost by model on screen.
Behind the call: Budgets nest: a month holds days, a day holds workflows, a workflow holds runs. Each budget names what happens at its limit: pause the run, downgrade to a fast model, switch to the pix allocation, escalate to a person, hard stop, alert only.
What you get: Spend you can hear in one sentence, capped where you decided.
From our own factory: 863.5M tokens in 30 days, $195 at list price.
Ask what it was worth
For example, you say: "What would this month have cost a team doing it by hand?"
Pix: Pix reads the same delivery figures the cockpit shows: the outcomes, the people and agent hours behind them, and what a team doing the same work by hand would have needed. It calls the by-hand side an estimate, because the rates come from your settings.
Behind the call: Delivery marks every figure measured, derived or assumed. Change a rate in Settings and every figure moves.
What you get: A return on the factory you can defend, with the working shown.
From our own factory: The same 421 outcomes took 9 hours of people time and $1,224 all-in with the factory. Our settings estimate a team by hand at 1,001 hours and $111,941, within a range of $54,416 to $115,925. That side is an estimate.
When something looks wrong, stop it
For example, you say: "Stop the patcher session on the payments repository."
Pix: Pix stops a session by voice: it reads back which one and stops it on a plain yes. The other four stops are one press each in the console on the same phone, for an owner or admin, behind a confirm.
Behind the call: Suspect an agent read something poisoned, a prompt injection in an issue for example? Throw the A2A blackout and Vulnetix stops relaying agent messages between hosts while you look, then review what the agents noted on the memory page. The kill switch stops every automation, and Stop machine destroys the sandbox container.
What you get: Every stop the factory has, from one session to the machine, in your pocket.
- Stop a session (By voice on a call): Ask Pix. It reads back which session it will stop and acts on a plain yes. Halts: One agent session, wherever it runs. Keeps working: Every other session, workflow and host.
- Pause all automations (One press in the cockpit): The red button at the top of the cockpit, with a reason for the ledger. Halts: Every trigger is held and no new step starts. Keeps working: Steps already running finish. Interactive Belai sessions are not touched.
- Stop the factory (One press in Settings): The factory kill switch at the top of Settings, with a reason for the ledger. Halts: Running steps are cancelled, board leases released, forge tokens revoked and pending wakes cancelled. Keeps working: Interactive Belai sessions keep working. Resume the factory is one press.
- Stop machine (One press on Pix Sandboxes): Stop machine on the Pix Sandboxes page. Halts: Belai stops and the sandbox container is destroyed, so nothing is left running. Running sessions end and uncommitted work in the clones is lost. Keeps working: Your repositories and everything already pushed. Start machine brings up a fresh one and clones the repositories again.
- A2A blackout (One press in the cockpit): Beside the pause in the cockpit, and in Settings, with a reason for the ledger. Halts: Vulnetix stops relaying agent broadcasts, direct messages and replies between hosts. Keeps working: Each host keeps coordinating its own sessions and logs every message. Memory notes are not part of the blackout: review or clear them on the memory page.
Every call stays on the record
For example, you say: "What did we agree on yesterday's call?"
Pix: Pix keeps notes you ask it to keep, and every call is kept: the words, the actions Pix took and the tables it read. Search past calls, download one as Markdown, or press Call Pix to pick it up where it ended.
Behind the call: The factory keeps its own record: the event stream, the run traces and an audit ledger whose entries are chained so the record can be verified.
What you get: Nothing done by phone is lost or unaccounted for. Audio is not recorded.
What Pix will not do
Pix runs on the same rules as the rest of the factory, and a few of them are written in code rather than left to the model.
- No action without a plain yes: Starting or stopping a session passes only when your very next answer is a plain yes. A tap on the read-back card is judged the same way.
- No standing permissions: Pix can allow one action once. It cannot allow an action for good, and plan reviews wait for the screen.
- No figures of its own: Every number it says comes from a tool reading your records. A total comes from the records, never from the model adding things up, and a figure not recorded yet is said as "not recorded yet", never as zero.
- No audio kept: Past calls keep the words, the actions and the tables, so you can search them, download them as Markdown or pick one up again. Audio is not recorded.
Put the factory in your pocket
A demo runs one Software Factory template on a single repository. You call Pix, hear what it did and what it cost, and try the stops yourself before anything widens.