Frequently asked questions.

Letco is an autonomous, quality-driven AI software development platform that builds production-ready software end-to-end while humans stay in control at every step. Unlike AI tools that stop at generating code, Letco independently verifies that what it built actually works and keeps it aligned with the original specification.

Yes — that is one of the two ways teams use Letco, and it is where most of them start. You can run the AI developer tools your people already use inside infrastructure you control, with every session recorded, and never touch the planning and verification pipeline. The autonomous side is there when you want it; it is not a precondition, and it is the same platform and the same subscription either way.

Letco is a platform for the whole software team, not just developers. Product owners, product managers, analysts, developers and testers all work from one living project documentation across the complete software development lifecycle (SDLC).

AI coding assistants are developer tools bolted onto an IDE that stop once code is generated. Letco is a full-team web platform that starts from a structured process, orchestrates the entire SDLC and independently verifies the result — solving “we can trust it works,” not just “the code got written.”

Every session is stored encrypted — the whole exchange with the model, not just the resulting diff. Each session has its own encryption key, and that key is wrapped for the organisation as well as for the user who ran it, so the record belongs to the company rather than to one developer's account and stays when that person moves on. Individual transcripts can be exported, and all of them can be backed up at once, including scheduled recurring backups into your own S3 bucket.

No. By default the AI runs on your own machine through the local agent, so your repository stays where it is — Letco's database holds project names, paths and metadata, not your code. If you choose a cloud session instead, the code lives in an isolated sandbox that is destroyed when the session ends. Prompts and code context travel directly from wherever the session runs to the AI provider; Letco does not proxy them.

Only the directories you explicitly register under a project. The agent authenticates with a revocable per-device token, and messaging runs over NATS with per-device scoped subjects, so a device can only ever see its own traffic. Its identity key is stored on disk encrypted with a key derived from the device token, and a compromised device can be revoked centrally at any time.

Yes, and at two levels. You can register your own local Kubernetes cluster so sandboxes execute on hardware you control instead of in Letco Cloud. For stricter requirements the entire platform can be deployed on-premise, in which case sessions, stored content and encryption keys never leave your environment.

Yes. Every AI session transcript can be exported individually, and all your conversation transcripts can be backed up at once — including scheduled, recurring backups to your own S3 bucket. Deleting your account force-completes active sessions and cascades deletion across your data, and project history can be erased on request.

Letco runs a closed four-stage pipeline — Analyze, Implement, Execute, Verify — with an automated repair loop. An adversarial “definition of done” auditor, separate from the code's author, checks every result against the documented requirements, and unresolved findings automatically spawn linked follow-ups. The project specification is pinned as a read-only source of truth and deterministically injected into every step, so the build stays aligned with the original intent instead of drifting as work moves along.

Missions are coming soon. You describe the goal in plain language, the mission agent asks clarifying questions and turns the answers into a task specification with success criteria and an intended approach, and then it waits. Nothing runs until you confirm, by design. Once started it drives sessions, planning and orchestration itself, reports back with a full trail, and can be paused, resumed or cancelled at any point; when it reaches a decision it should not make alone, it stops and asks.

Failure is a normal state in the pipeline, not an exception. Each stage closes with a gate: if something is missing or inconsistent, the stage reports findings instead of passing. After execution, an independent verification pipeline reviews the result — that it runs, plus quality, completeness, tests and documentation — and lists what it found. You can have findings repaired and re-checked automatically, or judge them yourself. In autonomous mode the orchestrator escalates at uncertain points and asks you to continue, clarify or abort rather than guessing.

Yes. A run can be scheduled into weekly time windows in your own timezone, so the work happens overnight or outside business hours. Windows gate the dispatch of new phases only — anything already running is never cut off mid-way. If AI usage approaches your provider's limit the run pauses itself and then resumes on its own once usage recovers and it is back inside a window; a pause you triggered by hand stays paused until you resume it.

Yes, at several points. Every pipeline stage closes with a gate you approve, and a supervised mode lets you launch each phase yourself instead of letting the orchestrator run to the end. After execution, the verification pipeline lists its findings, and the work is only closed when you confirm it with Accept & Complete.

Letco is multi-model: it orchestrates AI developer tools such as Claude Code, Gemini CLI, Codex CLI, Grok CLI and Antigravity, running Claude, Gemini, Codex and Grok. You can swap models per task, balancing cost against capability, without changing how you work.

You bring your own AI access. For Claude Code, the tool uses the login you already have on your own machine, and that credential never touches Letco. For other providers, keys you add are stored AES-256-GCM encrypted, injected only into your running sessions, and never returned by the API. Because the cost sits on flat subscription plans rather than per-token APIs, it stays predictable no matter how many tokens a run consumes. If your organisation cannot use developer subscriptions at all — procurement rules, or a policy against individual accounts — talk to us, because that case is handled separately.

Linux (Ubuntu, Debian, Linux Mint, Pop!_OS, Fedora, RHEL, CentOS, Rocky, AlmaLinux, openSUSE, SLES, Alpine), macOS, and Windows 10/11 including WSL2. The installer ships native binaries — no Node.js or build tools required. You register the device in the app, copy the one-line install command it gives you, and run it in your terminal.

Yes — that is the normal case. You register the directories the AI is allowed to work in, each holding a git repository you already have checked out, and the agent works against that local copy whatever the remote is. For cloud sessions, which clone the repository themselves, you connect GitHub, GitLab or Bitbucket; GitHub can also be connected as a GitHub App, which is the recommended route because no OAuth secrets change hands.

Git hosting — GitHub, GitLab and Bitbucket. Notifications go to an in-app centre and can escalate to email, Slack, Microsoft Teams or Discord, set per person or shared by a workspace. Jira works in both directions: Letco comments on the issue an escalation belongs to, and issues labelled for Letco arrive as work offers on the planning page for someone to accept or dismiss. For what the AI itself can reach, Context7 supplies current version-specific library documentation, and Claude Code plugins can be enabled per person from a catalogue your administrators curate.

About 30 minutes end to end: roughly 5 minutes for the account, 10 for installing and connecting the device agent, 5 for creating a project and registering its directories, and a couple of minutes for your first session. The full walkthrough with screenshots is on the Getting Started page.

Full platform access at no cost: the complete four-stage pipeline, automated verification and multi-model support. It also includes hands-on onboarding — a Letco engineer sets up your first project with you and stays with it until you have a shipped result. We take a small cohort of teams at a time. You still bring your own AI subscription; that cost sits with your provider, not with Letco.

Letco is in its pilot phase and pricing for general availability is not published yet. What is already settled is the shape of it: you keep paying your AI providers directly on your own plans, and Letco's fee is for the platform. Pilot teams get their pricing in writing before the pilot ends, with no obligation to continue — and if you decide not to, you can export every session transcript and your planning documents first. For a budget figure, talk to sales@letco.ai.