Skip to content
Akno

Getting started

This guide takes Akno from an empty installation to a useful, read-only first result. New recommended and model-free configurations leave knowledge-base files unchanged; updates and specialist configurations may retain explicit metadata or rendition write opt-ins.

Requirements

  • macOS;
  • Node 22.18 or newer;
  • a folder containing Markdown notes, optionally with attached documents.

Akno does not download or run models. Every model role is optional and points to an OpenAI-compatible endpoint you choose.

Install and configure

npm install -g @tenphi/akno
akno init

Guided setup first asks about configuration and write boundaries:

  1. Which folder is the knowledge base?
  2. Which model strategy should Akno use?
  3. Is it connected to a trusted agent, operated with human review, or intended to remain read-only?
  4. Which maintenance profile should be written?
  5. Which visible top-level folders are managed memory and which are source/reference material?
  6. Should unroutable durable claims have one exact managed fallback page?

Folder discovery reads directory names only. A selected managed folder becomes knowledge + remember: integrate; a selected source/reference folder becomes source + remember: deny. Every other discovered folder is written explicitly as searchable read-only knowledge, so its files remain intact but still participate in recall. Hidden folders and Akno’s operational directories are not offered. Existing configurations default to keeping their current folder and fallback policy untouched.

For an autonomous trusted-agent setup, the fallback offer defaults to memory/inbox. Accepting it adds an explicit memory/** managed rule when needed, but does not create the folder or page. Normal routing and an admitted model-suggested page still take priority; see Remember fallback.

It then shows the exact configuration overlay and a path-only diff before asking to write. Once the config is safe on disk, it separately offers to build the index, run a first recall, and install the macOS background service plus nightly schedule. Every follow-up explains its boundary and defaults to no. Declining one leaves an explicit command to run later; a failed follow-up never rolls back the valid config. Non-interactive setup always stops after the configuration write. Accepted actions stay pinned to the knowledge-base and state paths shown in the preview, including paths written into background-service definitions.

Indexing under the OpenAI preset sends configured page or document inputs to OpenAI to build derived search data. Model-free indexing stays local. A first recall is offered only after the index succeeds, and prompts for your actual question rather than running an irrelevant example query.

Choose a model strategy

StrategyUse it whenResult
OpenAI minimumYou want the simplest hosted setupOne Responses endpoint and credential; text-embedding-3-small for embeddings and gpt-5.6-luna for other roles
No modelsYou want local lexical retrieval onlyEvery model caller is explicitly disabled; existing provider definitions remain dormant
Specialist/manualYou already have role-specific endpointsAkno writes the path/profile and preserves provider and model blocks

For the OpenAI option, put the credential in the environment. The config stores only its variable name:

export AKNO_OPENAI_API_KEY="..."
akno init

The optional setup preflight sends only invented fixtures. It checks embedding access separately from Luna’s structured ranking contract, so access to one model cannot hide a denial for the other.

For automation, supply every required value explicitly:

akno init --preset openai-luna --akno-path /path/to/markdown \
  --maintenance autonomous --dry-run --check

akno init --preset no-model --akno-path /path/to/markdown \
  --maintenance audit

An existing config is never silently replaced. Guided updates default to leaving folder policy unchanged, show a path-only diff for every accepted change, and default to not applying it. Non-interactive updates require --force and do not classify folders or configure a fallback implicitly.

Before reviewing an existing installation interactively, inspect its authority-preserving proposal:

akno doctor --no-probe --admission-preview

It lists only top-level folder globs and counts—never page slugs or content—and proposes remember: deny so implicit read-only behavior becomes explicit without promoting any page. Root-level pages remain a separate count. Use akno init when you are ready to classify folders deliberately.

Build the first index

akno index
akno doctor

Run index before doctor: many diagnostics compare configured capabilities with what the index actually contains.

Indexing is read-only by default. It scans Markdown and documents, derives its search state, and leaves the set and bytes of knowledge-base files unchanged. The notable opt-ins are documented in Configuration.

Ask the first question

Use recall when you want inspectable evidence:

akno recall "How long is the Zephyr QX-100 warranty?"

Use answer when you want Akno to compose and verify the direct response:

akno answer "How long is the Zephyr QX-100 warranty?"

Every useful excerpt carries a Markdown line or document-page locator. Read the exact source when needed:

akno read products/zephyr-qx-100

See Reading memory for the difference between recall, answer, graph, timeline, and automatic context.

Try the demo knowledge base

The repository contains only invented material under examples/demo-brain.

cp -R examples/demo-brain /tmp/akno-demo
export AKNO_STATE_DIR=/tmp/akno-demo-state
akno --akno-path /tmp/akno-demo index
akno --akno-path /tmp/akno-demo doctor
akno --akno-path /tmp/akno-demo recall "Who services the boiler?"

Copy the demo before trying writes because remember, ingest, and enabled maintenance phases can change it.

Add background operation

After the first read-only session works, install the watcher without nightly maintenance:

akno service install --no-dream
akno service status

The service watches the folder, reconciles external edits, and gives hosts one long-lived process for the index and model clients. Install the nightly schedule only after reading The dream cycle:

akno service install

A safe adoption path

  1. Read only: use index, recall, answer, read, list, timeline, and graph.
  2. Explicit writes: try write or remember --dry-run; inspect the returned change id and test undo.
  3. Document intake: declare folder rules, ingest one invented or replaceable file, then consider an inbox.
  4. Watcher: install service --no-dream and verify external editor changes are reconciled.
  5. Maintenance audit: run akno dream --mode audit, inspect pending plans, and only then choose review or autonomous authority.

The sequence is about learning the trust boundaries, not satisfying a technical dependency.

Working from a checkout

pnpm install
pnpm build
cp config/local.example.jsonc config/local.jsonc
pnpm akno init
pnpm akno index

Use pnpm akno, not the installed akno, while developing. After code changes, pnpm akno redeploy builds, restarts the service, and waits for its socket. See Operations.

Search documentation