The one-sentence version#
Evodira turns what your people know into verified, citable organisational memory, without collecting your files. An agent runs on each machine, extracts and redacts on that machine, and sends typed claims and the redacted quotes behind them. A human approves every claim before anyone can get it back as an answer.
Two consequences fall out of that design, and most of this documentation is downstream of them. First, there is no central index of your raw documents, so “what does Evodira have?” is a question with a short, checkable answer. Second, nothing becomes answerable memory automatically. A queue of decisions is not a bottleneck we failed to remove; it is the product.
The path knowledge takes#
Every piece of knowledge in Evodira moves through the same five stages, in order. If you only remember one thing from this page, remember this sequence. Every surface in the console belongs to exactly one of these stages, and knowing which one tells you where to look.
| Stage | What happens | Where you do it |
|---|---|---|
| 1 · Capture | The local agent reads an approved folder, or a connector delivers an object. Extraction and redaction happen on the device, before any model sees the text. | Connectors, Evidence |
| 2 · Propose | The agent drafts typed claims (a decision, a runbook step, a relationship between a person and a system), each quoting the evidence it came from. | Ingestion |
| 3 · Decide | A human steward approves, rejects, or asks for changes. Approving makes a claim answerable. Publishing certifies it. | Review |
| 4 · Answer | Ask retrieves only from approved memory the asker is permitted to see, cites every sentence, and abstains when the evidence is thin. | Ask Evodira |
| 5 · Carry | The graph shows who knows what, risk scoring shows what would be hard to recover, and a departure case packages it for a successor. | Memory Map, At-risk, Handover |
The vocabulary, once#
Six words carry most of the product. They are used precisely and they are not interchangeable.
- Workspace
- The tenant. It has a region, a set of members with roles, one governance policy, and its own memory. Nothing crosses between workspaces. Also called an installation in permission names like installation.read.
- Evidence
- An immutable source object (a document, a message export, a transcript) with quote-level provenance. Evidence is what claims are quoted from. It carries its own access-control list.
- Claim
- One typed assertion drawn from evidence: a decision that was taken, a step in a runbook, a fact about a system. A claim is either drafted, in review, verified, or published. Only verified and published claims can be returned by Ask.
- Knowledge asset
- A group of claims that belong together (a runbook, a decision record, an operating note) with owners and a lifecycle state.
- Endpoint
- One machine running the Local Agent, enrolled against a deployment profile and holding a device-bound credential of its own. Also called a device or a node.
- Deployment profile
- The desired configuration a group of endpoints converges on: how machines join it, how much the agent may do unasked, what may leave a device, and which sources are pre-approved.
Where things are in the console#
The console is grouped into six zones, and the grouping is not cosmetic. It separates two audiences who share one application. An operator runs an estate of machines. A steward decides what becomes memory. Most people are one or the other, and the zone headings tell you which half of the product you are standing in.
- Operate
- The estate as it is right now: endpoints, health, drift, gateways, topology.
- Deploy
- What the estate is supposed to be, and how machines join it: profiles, provisioning, rollouts, control planes.
- AI runtime
- Which models the fleet is qualified to run, and where.
- Knowledge
- The other audience: review, memory map, handover, connectors, ask, assets, risk, evidence, expertise, transfers, ingestion.
- Governance
- Policy, automation, and the record of what happened.
- Workspace
- This installation, and your own place inside it: settings, your device, your contributions.
The sidebar is curated, not complete
The rail shows twelve destinations. There are twenty-six. The rest live on the console home page, grouped by zone, and every one of them is reachable from the command palette. Press ⌘K (or Ctrl K) anywhere in the console to jump, run an action, or search.
Which chapters you actually need#
Pick the shortest path that matches your job.
You are setting Evodira up for the first time
- Read chapters 2 through 8, in order. That is workspace → roles → profile → agents → sources → connectors → review. It is about ninety minutes of reading and it is the whole setup.
Someone installed Evodira on your machine
- Read chapter 6 (what it may read), chapter 9 (asking questions), and chapter 13 (your device and your controls). Skip everything else.
You have been made a steward or reviewer
- Read chapter 8 first, because the review queue is your entire job, then 9, 10 and 11.
You run the fleet
- Chapters 4, 5, 14 and 15: profiles, install, estate operations, and the commands that answer a support ticket.
You are done when
You can name the five stages in order, and say which console zone owns each one. Everything after this chapter assumes that sequence.
Something here wrong, missing, or no longer true of the product? Write to hello@evodira.com. Documentation that has drifted from the console is a bug and we treat it as one.