Question 1

How do I sign up?

Accepted Answer

Go to sami-agent.com and click "Get started". You can register with an email address or continue with GitHub or Google. After registration you receive a verification email — click the link and your account is active.

Question 2

How do I install the desktop app?

Accepted Answer

Visit /download and grab the installer for your platform (Windows, macOS, or Linux). Run the installer, then sign in with your SAMI account. The app keeps itself updated in the background.

Question 3

What is a consensus run and how do I start one?

Accepted Answer

A consensus run is a structured task session where an AI council works through your request in stages — gathering context, planning, drafting, reviewing, and finalising — before producing a result. Open the desktop app, describe your task in the War Room, and click Submit. The council begins immediately.

Question 4

How do I set up my workspace?

Accepted Answer

In the desktop app, open a folder from your file system using the Explorer panel or File → Open Folder. SAMI reads the project structure and makes it available to the council as context. You can open multiple folders as a multi-root workspace.

Question 5

What verbosity modes are available?

Accepted Answer

Three modes: Silent (final output only), Normal (stage summaries + key decisions), and Verbose (every stage turn visible in the conversation pane). Switch at any time from the toolbar dropdown — the change applies to the next run.

Question 6

What are approvals and do I need to click through them?

Accepted Answer

Approvals are checkpoints where the council pauses and asks you to confirm before writing files, running commands, or making significant changes. On the default autonomy level most positive-vote stages advance automatically; only destructive or irreversible actions require an explicit click. You can raise or lower the autonomy level in Settings.

Question 7

Does SAMI work offline?

Accepted Answer

The desktop app requires an internet connection to reach the AI inference infrastructure. Local editing, file browsing, and terminal use work offline, but no council run can start without a network connection.

Question 8

Do I need to configure anything before my first run?

Accepted Answer

No. After installing and signing in, you can start a consensus run immediately. The model catalog, credit allocation, and team settings are all pre-configured based on your plan. Advanced users can fine-tune the council size, autonomy level, and verbosity in Settings.

Question 9

What is S.A.M.I.?

Accepted Answer

S.A.M.I. (Strategic · Agentic · Multilayer · Intelligence) is a desktop coding environment that runs an AI council instead of a single model. Multiple AI models propose, critique, vote, and converge on a result before any change lands in your workspace.

Question 10

What is S.A.M.I. not?

Accepted Answer

SAMI is not a chatbot, a prompt playground, or a single-model autocomplete tool. It is not a code generator that pastes a snippet and walks away. Every task goes through a structured pipeline where independent AI perspectives check each other before you see a result.

Question 11

How does SAMI differ from single-model coding assistants?

Accepted Answer

Single-model tools get one shot. If the model makes a flawed assumption or misunderstands scope, you inherit that mistake. SAMI's council architecture means the first draft is challenged by other models in a structured review stage — logical errors, missing edge cases, and scope creep get flagged before you see the result. The tradeoff is more compute per run; the benefit is fewer runs that need a full redo.

Question 12

What does a consensus run actually look like end-to-end?

Accepted Answer

A run passes through seven named stages: (1) Connector — an interactive Q&A loop where the council gathers any missing context before work begins. (2) Strategist — a structured plan is produced and checked for completeness. (3) Architect — system design, interfaces, and contracts are defined. (4) Maverick — the council produces the draft implementation. (5) Optimizer — the draft is refined and challenged for correctness. (6) Guardian — a red-team and security audit is performed; a blocking no must carry a written rationale, and the council deliberates until the objection resolves into a genuine yes. (7) Debugger — the final validation pass before the result is written to your workspace.

Question 13

Who is SAMI built for?

Accepted Answer

Primarily professional software engineers and engineering teams who need reliable AI-assisted development — not just code generation but structured reasoning about their specific codebase. SAMI is also useful for technical leads reviewing complex changes, researchers prototyping software experiments, and companies in regulated industries that need an auditable AI workflow.

Question 14

Which languages and frameworks does SAMI support?

Accepted Answer

SAMI is language-agnostic. The council reasons in natural language about your code, so any language or framework that can be represented as text is supported. Python, TypeScript, Go, Rust, Java, C#, SQL, Terraform, and dozens of others have all been used successfully.

Question 15

Is SAMI a code editor?

Accepted Answer

The desktop app includes a full code editor so you can view, edit, and review files alongside the AI conversation. It is not a stripped-down viewer — you get syntax highlighting, file tree, integrated terminal, and git integration. The editor is the AI work surface; managing your account and billing happens in the web dashboard.

Question 16

Does the AI run on my machine or in the cloud?

Accepted Answer

AI inference runs on managed cloud infrastructure — not on your machine. Your code and context are sent to the AI council, the council responds, and nothing is stored server-side after the session. Sensitive material never leaves the encrypted request/response channel.

Question 17

Which operating systems are supported?

Accepted Answer

The desktop app runs on Windows (x64 and ARM64), macOS (x64 and Apple Silicon), and Linux (x64 and ARM64). The web dashboard works in any modern browser — Chrome, Firefox, Safari, and Edge.

Question 18

What is the web dashboard for if the AI is in the desktop app?

Accepted Answer

The web dashboard handles everything account-related: plan management, billing, seat provisioning, usage analytics, and team administration. The separation is deliberate — AI workloads belong in an environment with direct file system access and a proper editor; account and billing belong in a secure, browser-accessible backoffice.

Question 19

Is there a mobile app?

Accepted Answer

Not currently. The desktop app is the AI work surface and needs a real file system and editor context to be useful. The web dashboard is mobile-responsive for checking usage or managing your account on the go.

Question 20

Does SAMI remember earlier work on the same project?

Accepted Answer

Yes. Context is not reset between tasks. A follow-up run on the same project carries forward the decisions and lessons from earlier runs, so the council does not re-litigate a choice you already settled or repeat a mistake it already corrected.

Question 21

What does the name S.A.M.I. stand for?

Accepted Answer

Strategic · Agentic · Multilayer · Intelligence. Each word describes a design principle: the council forms strategy before acting, each model in the council is an autonomous agent, the pipeline is multilayered with independent review stages, and intelligence emerges from the interaction rather than from a single model.

Question 22

What is the War Room?

Accepted Answer

The War Room is the main interface in the desktop app where you interact with the AI council. You describe your task, attach files or context, set the verbosity and autonomy level, and submit. The council's staged pipeline runs in the conversation pane and writes results directly to your workspace.

Question 23

What stages does the pipeline run through?

Accepted Answer

The pipeline has seven named stages: 01 Connector (interactive context Q&A), 02 Strategist (planning), 03 Architect (system design and contracts), 04 Maverick (draft implementation), 05 Optimizer (refinement and challenge), 06 Guardian (red-team and security audit), 07 Debugger (final validation). Each stage leaves a visible artifact and vote record.

Question 24

What is a veto round and how is it resolved?

Accepted Answer

A veto round is a review stage where any Guardian can cast a blocking no vote. A no must carry a written reason — at least three sentences and one concrete piece of evidence. An objection is raised as a no with a written reason, and the council resolves it into a genuine yes through discussion. When a no lands, the stage pauses and the full council re-deliberates with the rationale surfaced to all Guardians. The council discusses counter-arguments and alternatives until the objection resolves into a genuine yes or the no is upheld. If the no is upheld, the run advances with an explicit risk note attached to the final artifact, visible in the War Room. Every vote is yes or no — a no always carries its reasoning, which is what drives the resolution loop.

Question 25

Can a review stage send work back to the design, not just the code?

Accepted Answer

Yes. When a reviewing stage finds a code-level issue, the change goes back to the draft. When the issue is structural — the interface or module boundary itself is wrong — the objection routes further back to the Architect stage with a concrete, written change order the council must resolve before the draft is touched again. Either way, the loop is bounded by your run budget, so refinement converges instead of spinning indefinitely.

Question 26

Does the council actually run my project, or just review the code?

Accepted Answer

Both. The draft and validation stages don't just produce an opinion — the council delivers a complete project structure in one pass, scaffolding and logic together, then starts the project and operates the terminal the way a developer would: install, build, test, lint, typecheck. Errors come back file by file, not as a vague summary, so the fix that follows targets the actual failure.

Question 27

What is the Execution-Verify Gate and how does it differ from a normal vote?

Accepted Answer

Most AI tools vote on plausibility — they read the code and decide if it looks right. SAMI's Execution-Verify Gate makes the council vote on proven behaviour: it actually runs the project's build, tests, lint, and typecheck and feeds the real output back to the council before any verdict is reached. A passing run is proven, not assumed. A failing run is placed in front of the council with the real failure visible — the council either accepts it (a known, expected, or flaky failure, with the decision recorded) or treats it as a blocker. A blocker flows into the same never-abort resolution loop as any no: the council deliberates with the real failure in view until it resolves into a genuine yes; only run-budget exhaustion or cancelling advances with the risk noted. The run is never silently stopped.

Question 28

Who decides the final answer in a run?

Accepted Answer

The council votes at each stage. The protocol is binary: yes or no. A no is always explained and always resolved before a stage advances. The final output is the artifact that clears every stage with every no resolved into a genuine yes.

Question 29

How do I read the War Room conversation?

Accepted Answer

Each message in the conversation pane is labelled with the stage it came from and the council voice that produced it. Stage labels appear as collapsible headers; expand one to see the full reasoning. In Silent mode you only see the final output; in Verbose mode every intermediate thought is visible. Use the stage progress bar at the top to track which phase the current run is in.

Question 30

Can I interrupt or cancel a run mid-way?

Accepted Answer

Yes. Click Stop in the toolbar at any time. The run is cancelled cleanly — partial workspace changes are not written unless a stage already completed and applied its output. If you want to resume from a specific point, start a new run and describe what was already done; the council will pick up from your description.

Question 31

How long does a typical run take?

Accepted Answer

It depends on task complexity and the number of council members active on your plan. Focused bug-fix or explanation runs finish quickly; a multi-file refactor or architecture proposal with full review rounds takes longer. Token consumption per run is visible on your usage analytics page.

Question 32

Can I attach files or images to a run?

Accepted Answer

Yes. You can attach files from your workspace by clicking the paperclip icon in the War Room toolbar. Images (for UI or diagram tasks), text files, and code snippets are all supported. The council treats attachments as additional context alongside your workspace structure.

Question 33

Which AI models can I choose from?

Accepted Answer

The model catalog is curated and updated as providers release new versions. It includes flagship models from major providers across reasoning, coding, and instruction-following specialisations. Specific model IDs are not listed publicly because the catalog rotates — what you see in the app is always current. You pick models in the desktop app's model picker.

Question 34

Why use a council of models instead of just one?

Accepted Answer

A single model gets one shot at your task. It cannot internally challenge its own reasoning or catch its own misunderstandings. A council with independent voices and structured review stages is architecturally immune to a class of single-model failure: overconfident wrong answers, tunnel-vision on one approach, and silent scope creep. The cost is more tokens per run; the benefit is results that have already been cross-examined.

Question 35

How do I choose which models go in the council?

Accepted Answer

Open the model picker in the desktop app. Every model you select becomes a council Guardian — an equal voting participant across every stage. Two of those Guardians take on per-stage coordination duties (one coordinates the stage, one writes the stage artifact), but all cast votes with equal weight. For most tasks, using SAMI's recommended council preset is the best starting point — it balances reasoning quality and cost.

Question 36

What roles do models play in the council?

Accepted Answer

Every model you add to the council is a Guardian — it participates in every stage with equal voting weight. There are no separate 'review models' or 'draft models'. Per stage, the council assigns two coordination duties on top of the Guardian role: one model acts as Lead (coordinates the stage flow and aggregates votes) and one acts as Writer (produces the stage artifact). These are temporary per-stage responsibilities, not fixed roles — the same model that writes in one stage may coordinate in the next. A single blocking no vote from any Guardian vetoes the stage regardless of which coordination duty that Guardian holds.

Question 37

How many models can be in the council on each plan?

Accepted Answer

Council size depends on your plan — check the pricing page for current limits. Higher plans unlock larger councils, which means more independent review passes per run. Enterprise can configure council size on request.

Question 38

What happens on the Free plan when it comes to model access?

Accepted Answer

Free plan users get access to a selected subset of the catalog — capable models optimised for cost. You will not have access to the highest-capability or highest-cost models. Upgrading to Pro or higher expands the catalog and council size available to you.

Question 39

Can I use my own provider API keys (BYOK)?

Accepted Answer

BYOK (Bring Your Own Keys) is available exclusively on the Enterprise plan, for organisations that need to route inference through pre-approved provider contracts or compliance-specific endpoints. On Free, Pro, Team, and Ultra, all inference runs through the managed catalog — no external keys are needed or accepted.

Question 40

Does output quality differ between plans?

Accepted Answer

Yes, in two ways. First, higher plans unlock higher-capability models. Second, larger council sizes mean more independent review passes, which statistically catches more errors — so a run on a higher plan generally produces stronger results than the same task on the smaller Free mini-council.

Question 41

Can I use local models or Ollama?

Accepted Answer

No. SAMI does not support local model execution or Ollama on any plan other than Enterprise BYOK. All inference runs through managed cloud infrastructure. This ensures consistent performance, no GPU required on your machine, and uniform security guarantees across all users.

Question 42

What is the scribe model and how is it different?

Accepted Answer

After a run completes, SAMI generates a plain-English summary of what changed and why. On most plans this is automatic; higher-tier plans can configure the model used for this summary.

Question 43

What happens if a model I rely on is retired by its provider?

Accepted Answer

SAMI monitors provider model lifecycle announcements and automatically routes retired models to their approved successors. You will see a brief notification in the desktop app the first time a substitution is made, and your runs will continue uninterrupted. You don't need to do anything.

Question 44

What plans are available and what do they cost?

Accepted Answer

Five plans: Free, Pro, Team (per-seat), Ultra, and Enterprise (custom pricing). Current prices and feature details are on the pricing page.

Question 45

What is a token quota and what counts toward it?

Accepted Answer

Every AI council run consumes tokens — the units of text that models process. Tokens are counted for both the context you send (your code, instructions, workspace) and the responses the council generates. The quota is the monthly ceiling for your plan. Usage over the quota is either blocked (on Free) or metered at an overage rate (on paid plans).

Question 46

What happens when I hit my quota?

Accepted Answer

When your credits are exhausted, new AI runs pause — nothing is charged behind your back. On Free, Pro, and Team you can purchase a credit top-up or upgrade to continue immediately, or wait for the next billing cycle. Ultra and Enterprise have no credit ceiling; every plan can additionally set a self-imposed monthly spending cap in billing settings, which pauses runs when reached.

Question 47

What payment methods do you accept?

Accepted Answer

Credit and debit cards, SEPA direct debit, Apple Pay, Google Pay, and selected regional methods through Stripe. EU customers on Team and Enterprise can also pay by invoice.

Question 48

Can I cancel anytime?

Accepted Answer

Yes. Cancel from the billing page in your dashboard. Your plan stays active until the end of the current billing period; no prorated refund is issued for the remaining days. After cancellation you revert to the Free plan.

Question 49

Do you offer refunds?

Accepted Answer

Yes within 14 days of the first payment for new subscriptions, and case-by-case for billing anomalies (double-charges, technical failures). Open a request via the contact form with your invoice number and the reason.

Question 50

How does Team plan billing work?

Accepted Answer

Team is billed per seat per month. The workspace owner is the billing contact. Seats are added through the team management panel; each new seat is charged pro-rata for the remaining days in the current billing cycle.

Question 51

Is there a free trial for paid plans?

Accepted Answer

The Free plan is a permanent free tier, not a trial. It runs on the same council infrastructure with a smaller council and a smaller token quota. You can upgrade to a paid plan at any time from your billing page. Refer to the pricing page for the current Free plan limits.

Question 52

Do you charge VAT or sales tax?

Accepted Answer

Tax handling depends on your billing country. EU business customers with a valid VAT ID can enter it in billing settings to receive invoices with reverse-charge VAT. Consumer accounts in applicable jurisdictions will see local VAT added at checkout. Stripe handles the tax calculation.

Question 53

Where can I view and download invoices?

Accepted Answer

Invoices are available on the billing page in your dashboard under the Invoices tab. Each invoice links to the Stripe-hosted PDF. If you need a custom invoice format for your accounting system, contact support with your invoice number.

Question 54

Do you train AI models on my code?

Accepted Answer

No. SAMI does not use your prompts, code, or outputs to train any model. Where the underlying provider offers a no-training endpoint, that endpoint is used. This is confirmed in our AI Use Policy.

Question 55

Is my chat history stored on your servers?

Accepted Answer

No. Conversation content streams through the backend and is not written to a remote conversation database. Chat history visible in the desktop app is stored locally on your machine. The web dashboard stores account, billing, team, and usage metadata needed to operate the service.

Question 56

Where does my code go when I run a task?

Accepted Answer

The code and context you explicitly include in a task is sent to the AI council over an encrypted channel. It is processed ephemerally for inference and the response is streamed back. SAMI does not keep a permanent server-side copy of your code or conversation content.

Question 57

Who are your sub-processors?

Accepted Answer

The complete list of third-party services that may process user data is maintained at /legal/subprocessors. Optional sub-processors (AI providers) only receive data when you actively select that provider's model.

Question 58

What are my rights under GDPR?

Accepted Answer

If you are in the EU or UK you have rights to access (Art. 15), rectification (Art. 16), erasure (Art. 17), restriction (Art. 18), portability (Art. 20), and objection (Art. 21). To exercise any of these rights email privacy@sami-agent.com. We respond within 30 days.

Question 59

Can I delete my account and all my data?

Accepted Answer

Yes. Go to Settings → Account in your dashboard and click Delete Account. This removes your account record, associated tokens, and billing references. Some Stripe records are retained for the minimum period required by applicable tax law. Since chat history is stored locally, deleting your account does not remove local files — you must do that yourself.

Question 60

What does "local-first" mean for my data?

Accepted Answer

Local-first means the data that matters most to daily work - conversation history, workspace state, and local IDE settings - is written to your machine by default, not synced to a remote chat database. SAMI's server only sees what you explicitly send in a task request plus the account and usage records required to run the service.

Question 61

Is there an audit log of who accessed what?

Accepted Answer

Admin accounts have access to an audit log of administrative actions in the admin panel. Individual user sessions are not logged at the message level because chat content is not stored server-side. Billing and account-change events are recorded and accessible in your billing history.

Question 62

Where is data physically stored?

Accepted Answer

Account data (email, plan, billing references) is stored in a managed cloud database in the EU. No chat content is stored server-side. Sub-processors may process data in their own regions as disclosed in the subprocessors list.

Question 63

Can I export my account data?

Accepted Answer

You can request a data export under GDPR Art. 20 by emailing privacy@sami-agent.com. We will provide the server-side data we hold about you (account record, billing history) in a machine-readable format within 30 days. Local files are already on your machine.

Question 64

How do I report a security vulnerability?

Accepted Answer

Email security@sami-agent.com or follow /.well-known/security.txt for our disclosure policy. Please do not post vulnerabilities publicly before we have had a chance to investigate and patch.

Frequently asked questions

Getting started

How do I install the desktop app?

What is a consensus run and how do I start one?

How do I set up my workspace?

What verbosity modes are available?

What are approvals and do I need to click through them?

Does SAMI work offline?

Do I need to configure anything before my first run?

Question not answered?

Frequently asked questions

Getting started

How do I install the desktop app?

What is a consensus run and how do I start one?

How do I set up my workspace?

What verbosity modes are available?

What are approvals and do I need to click through them?

Does SAMI work offline?

Do I need to configure anything before my first run?

Question not answered?