For AI agents: the complete documentation index is available at https://docs.flezise.com/llms.txt, the full documentation bundle is available at https://docs.flezise.com/llms-full.txt, and this page is available as Markdown at https://docs.flezise.com/getting-started/what-is-flezise.md.
Docs

What is Flezise

Flezise is an agentic software factory. One unified harness runs every stage of the software development life cycle, from the first stakeholder interview to the pull request, and people hold the gates between the stages.

One pass through the line: intent, the lifecycle documents, the approval gate, the board, and a delivered change.

Agents carry the work between stations. They interview stakeholders, draft the brief and the PRD, design the screens, write the code and the tests, review each other's output, and open the pull request. People decide what gets built, set the limits, and sign off on what ships. Every document is versioned, every run leaves evidence behind, and nothing advances to the next stage without the approvals you configured.

Why not a dark factory

Once you industrialize software delivery, two factory models appear. Almost no one reads most of the generated code any more, so what separates them is whether a person still governs the system and owns the decision to ship. Flezise runs the first kind, with the lights on.

Level 3 · where Flezise runs

Safe factory

Agents generate at volume; a person governs the system and owns the acceptance gate. The engineer stops reading lines and starts judging the change as a whole: does the evidence hold, does it solve the problem, is it safe to ship.

  • Deterministic checks catch syntax, types and regressions
  • People judge by evidence and spot-check by risk
  • A named person authorizes every change into production

Level 4 · experimental

Dark factory

No person governs any single change. Quality becomes a statistic across thousands of runs against service clones. Fast, and impressive. It also opens gaps nobody has closed.

  • No clear answer for who is liable when it fails
  • Nothing a procurement or audit team knows how to sign off
  • Oversight thins to auditing outcomes, so drift accrues quietly

What Flezise is good at

  1. Pinning down what to build before anyone builds it
    • Stakeholder interviews in one shared session, led by an agent
    • A coverage matrix of what is still ambiguous, and who answered what
    • Disagreements between contributors flagged and routed to an owner
    • The brief and the PRD drafted from the answers, versioned, approved by people
  2. Covering every stage, from requirements to release
    • Requirements planning: a brief and a PRD stakeholders sign off
    • User design: screens on a canvas, with variants and style directions, reworked from a plain-language comment
    • Construction: one approved spec at a time into a branch, tests written alongside
    • Review and testing on every increment
    • Cutover: a pull request with its evidence attached
    • Architecture diagrams as code; exports to PDF, Markdown, PNG or PowerPoint
  3. Keeping agents useful without letting them loose
    • A throwaway sandbox per task, which can be cut off from the internet
    • Automated checks after every change: build, types, lint, tests, secret scan
    • Failed checks go back to the agent for a bounded number of repair rounds
    • A verifier agent reviews every draft before a person reads it
    • A run that cannot converge pauses and asks
  4. Working on the code you already have
    • An existing repository indexed into a code graph
    • An as-is wiki of the architecture it found
    • Change specs raised against the current system
    • Drift detection when the code and the approved design part ways

Privacy, security and extensibility

Runs where your data lives

Self-hosted on a single host or on your Kubernetes cluster. Documents, code, embeddings and evidence stay in your own database and storage.

Bring the model providers you already trust, or a private endpoint, and assign a model to each type of work.

Locked down by default

Single sign-on, roles and approval policies decide who sees and signs what. Agents work in throwaway sandboxes that can be cut off from the internet, and provider keys never reach them.

Every model call is screened for prompt injection, PII and secrets. Budgets and rate limits apply per project.

Evidence an auditor can check

A tamper-evident trail of every action, including the ones an agent takes on your behalf. A signed evidence bundle per run that can be verified offline.

Usage and cost, per project and per organization.

Extends into your toolchain

Reusable packs of skills, templates and domain knowledge. Integrations with GitHub, GitLab, Bitbucket, Jira, Linear, Slack, Teams, Confluence, Notion and Microsoft 365.

A CLI and an API for automation, and project context served to the agents you already run.

Next steps