What this thing is, what it does, and exactly what you get for your money.
VIGIL is a private AI chat platform, built and run by hand. Under the hood it uses Z.ai's GLM models. The thinking is theirs; everything around it is ours: the chat, the projects, the document tools, and a sandbox for running generated scripts called VIGIL Studio.
The assistant is called SIERRA. Tell her your name once and she remembers it in ordinary chats and standard projects, until you delete it. A Job Hunter does not use that account memory: it follows the instructions, memory and files saved on that hunt. No ads, no data sales. Prompts go to Z.ai to generate replies. We do not train on your chats. Z.ai processes them under its own policy.
Available on Light and Pro. Create a Job Hunter project, set its search instructions, and upload a résumé or CV. SIERRA matches roles from that hunt only: its name, instructions, memory and uploads, plus what you say in its chat. Account memory and your other projects are not used. Ask SIERRA to research suitable roles and explain the evidence for each match. Open the original posting to check that it is current.
Document preparation uses your plan's credits. Failed generation is not charged. Job Hunter has no send button, mailbox connection or automatic application path. It does not bypass restricted job boards. Generated text can still be wrong: confirm every claim and follow the employer's instructions. There is no promise of a hiring outcome or passing an AI detector. Historical drafts remain available in existing hunts and your account export.
Chat that remembers you
Streaming conversations with full history. SIERRA keeps an account-wide memory of who you are and how you want to be addressed. Settings has a page called What SIERRA knows about you: every learned fact in one list, where you can add, delete, or export the lot as a file. Nothing hidden.
Projects, with a persona each
Workspaces with their own instructions, knowledge files, and memories. Set the project persona once (role, tone, format, rules) and every chat inside follows it, with one-click presets if you want a head start. Upload text, markdown, PDF or DOCX files: when you ask something, SIERRA reads the tree and pulls in only the files that match the question, so big projects stay fast and answers stay relevant.
Documents, properly formatted
Ask for a document and you get a real one: a themed PDF or Word file with a cover banner, contents page, styled tables and checklists. Drop your brand guidelines into a project and every document comes out in your colours. What you preview is what you download.
VIGIL Studio
When a template is not enough, SIERRA writes a Python script and the platform runs it in a locked-down sandbox. No internet, hard resource limits, everything audited. The finished PDF appears right in the chat.
Sharing
Share any conversation with a public link that expires, anywhere from an hour to thirty days, or kill it any time. Whoever opens it gets a read-only view, no account needed.
A persona with a spine
SIERRA is sharp, calm, and honest rather than agreeable. Think Cortana without the attitude problem. She keeps it clean, treats the platform's Christian values with respect, and will tell you when something is a bad idea.
Daily email briefs
Pick a topic and a time of day in Settings. Every day, SIERRA searches the web and emails you a short brief with sources: news, a market, a team, anything you follow. Each digest uses provider cost plus 2% and search fees, capped at the allowance shown in Billing. Failed email delivery is not charged.
An API you can build on
Create a key in Settings, point your code at the chat-completions endpoint, and VIGIL answers in the standard OpenAI shape. Keys are stored hashed and can be revoked any time. Usage bills from your credit balance, same as chatting.
Pictures, and pictures of your pictures
Generate images in square, portrait, or landscape, then hit the sparkle button for a fresh take on the same idea. Image generation is on Light and Pro. On every plan you can attach a photo and change its background to white, a colour, a blur or a cutout, free, on our server. Natural lighting and marked edits are on Light and Pro.
Your browser talks to VIGIL over HTTPS. An ordinary chat picks up your account memory and any project context. A Job Hunter leaves account memory and other projects out. The message passes through the SIERRA layer where her persona and skills are applied, and gets answered by a GLM model. Replies stream in as they are written. Studio scripts go one step further, into a container with no network, a read-only filesystem, and hard caps on memory, CPU, and time.
You pay in VIGIL credits. Chat uses each model’s provider input, cached-input and output prices plus 2%, converted at 500,000 credits per US dollar. Studio adds a separate execution fee shown in Billing. The conversation counter shows credits spent.
Build on VIGIL from your own code. Create a key in Settings → API Keys, then call the OpenAI-compatible chat-completions endpoint. It is non-streaming and billed from your credit balance at exactly what the model used plus 2%, 20 requests per minute per account.
curl https://chat.vigilglm.ai/api/v1/chat/completions \
-H "Authorization: Bearer vig_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3-flash",
"messages": [{"role": "user", "content": "Hello!"}]
}'You can chat without an account on the web: 30 messages every 5 hours on the free Flash model, 10 in a minute, with GLM-4.6V-Flash for image questions. Nothing is saved. An account adds saved chats, memories and the tools below. The Android, desktop and VIGIL-Code apps require an account.
VIGIL credits measure cost, while model tokens measure text or audio usage. Provider charges convert to credits with 2% markup. Different models and tools have different prices. Missing provider usage is marked estimated; platform execution fees are separate.
| Feature | How it bills | Credit cost |
|---|---|---|
| Chat messages | Model-specific provider input, cached-input and output cost, plus 2% | Provider USD × 1.02 × 500,000 |
| Web search | Per search that actually runs. Asking is free | 5,100 |
| Image generation | Model-specific provider image price, plus 2% | 5,100 fast / 7,650 premium |
| Image edit (reimagine) | Vision prompt at metered rates, plus the flat image price | Model cost + image fee |
| Background removal | Runs on our server. Never touches the model | 0 |
| Meeting transcript | Provider audio-token usage at $0.03 per million, plus 2%; missing usage is estimated from PCM duration | Provider USD × 1.02 × 500,000 |
| VIGIL Studio (code runs) | Platform execution fee, plus correction model usage. See Billing for current fees | See Billing |
| Documents, projects, memory, voice | No separate charge for local tools. Hosted voice remains subject to availability and usage limits | 0 |
| Plan | Price | Credits / month | Messages | Credit limits (5 hours / week) | Models | Studio |
|---|---|---|---|---|---|---|
| Free | $0 | 1,000,000 | 50 per 5 hours | 46,667 / 233,333 | VIGIL Flash | 3 runs an hour |
| Light | $5 / month | 1,500,000 | No cap | 70,000 / 350,000 | VIGIL Flash | 10 runs an hour |
| Pro | $19 / month | 6,000,000 | No cap | 280,000 / 1,400,000 | All + GLM-5.3 reasoning | 10 runs an hour |
Every plan has two credit limits: a week is 7/30 of the monthly credits and 5 hours is a fifth of a week. The 5-hour window opens with your first message, and your week runs on its own 7-day cycle from when your subscription started (or your signup). At a limit, chat, voice and VIGIL-Code keep working on GLM-5.3 Flash with Low thinking until it resets, while images, Studio, research, document parsing, Job Hunter and digests pause, and API requests get HTTP 429. The usage page shows both limits and when they reset. Free accounts are topped up to their allowance on the 1st of each month. Paid credits arrive when each monthly invoice is paid. Web search, the isolated browser, meeting transcription, document parsing, slides and Studio are on every plan and bill from your balance. Image generation, photo editing and Job Hunter need Light or Pro. Cancel whenever you like and you keep unused credits within your plan limits, and when a subscription ends everything drops back to free-tier levels. Each conversation keeps the most recent 32K tokens in view on Free and 64K on Light and Pro. We do not do refunds. All purchases are final, and the reasons why are in the Terms. Subscribing means you agree to them.
Studio exists because templates only get you so far. Ask for a bespoke one-pager and SIERRA writes a self-contained ReportLab script. One click runs it, and the PDF shows up in the chat, ready to view or download, with its exact cost printed underneath.
What the sandbox guarantees: no network access at all, a read-only filesystem, no root, no Linux capabilities, 512MB memory, one CPU, 64 processes max, a 45-second kill switch, and a 25MB file limit. Only PDF, DOCX, and PNG files come back, nothing bigger than 10MB. Every run has the execution fee shown in Billing, plus any model corrections, and is logged for security. Studio is on every plan: 3 runs an hour on Free and 10 on Light and Pro.
Your conversations are yours. They are never sold and never advertised against. We do not train on them. Prompts still go to Z.ai to generate replies, under Z.ai's policy. You can delete any conversation, project, file, or memory yourself at any time, and if you want your whole account gone, delete it in Settings or email us. The details are in the Privacy Policy, and the rules are in the Usage Policy.