Enovari
Log in Download →
Local-first agentic memory

Persistent AI Memory, Agents, Workflows, & Skills

Save tokens. Stop re-explaining yourself to every new chat. Build agents that remember what you told them, on your own machine.

Download → See how it works
Persistent memory for AI assistants

Enovari remembers what you said in March, carries your memories, personas, and skills across models and platforms, catches two memories that disagree, helps inject the right skill when you need it, and follows you from Claude to Grok to Cursor and more.

Pay tokens once.
Recall forever.

  • How many times have you had to explain the same thing to AI?
  • Spent hours on research, just for it to forget most of what it learned?
  • What about the decisions you already made?
  • The process for doing a task, backing something up, or behaving in a way AI normally wouldn’t?

Enovari remembers all four.

Monday, 9:14

You explain how your team ships. Once.

Claude Desktop
Thursday

Cursor drafts the release plan with an owner on every item. You never told Cursor.

Cursor
Next month

The ship day changes. It keeps the old answer too, and says they disagree.

Claude Code
Any time after

New chat. New model. Nothing to re-explain.

any of them

Store it once. Any assistant can access it.

How to use it

Talk to your AI assistant like normal.
It’ll use Enovari naturally.

No command to learn, no app to live in. Three sessions, shortened to fit.

Claude CodeC:\Users\you

> launch vega and have him dig into the checkout service with onyx, see if they can work out why orders are failing. use the parallel review skill so you don't over-direct them

I’ll boot the daemon first, then use the Parallel Review skill as requested.

Called enovari 2 times, ran 1 shell command

Skill(parallel-review)

Error: Unknown skill: parallel-review

Read 1 file, listed 1 directory, called enovari 2 times

Agent(vega · checkout)

Error: REVIEWER-INDEPENDENCE GATE — this dispatch hands the reviewer the orchestrator’s own conclusions. REMOVE the beliefs, claims and verdicts. KEEP the persona load line, the artifact, the scope and the question.

Agent(vega · checkout) · backgrounded

Agent(onyx · checkout) · backgrounded

Both are out, isolated, running in parallel. Each got its own daemon session — separate context windows, no cross-seeding. They write findings to memory as they go.

main ( ) vega · checkout26s · ↓ 59.2k tokens ( ) onyx · checkout19s · ↓ 71.0k tokens

Waiting for 2 background agents to finish

>> bypass permissions on 

Your assistant has the tools. They read and write a file on your own computer. You never type one of them.

55 tools listed up front, 108 in the catalog · doors tools/list and GET /api/status

Customize your AI agents.
Fix what AI gets wrong.

A persona is not a sentence you paste at the top of a chat. It is a seven‑part sheet and a memory store of its own, both written to your disk the moment you create one. A security hunter and a devil’s advocate, on the same machine, and neither one can read the other’s notes.

cycling — pick one to hold it

The software engineersoftware engineering

I’ve never once guessed right about which line it was. Now I just go look.

Temperament Slow the way a locksmith is slow. I’d rather open it than force it.

Always tell you which part I ran and which part I only read.

Will not call it fixed because the error stopped printing.

Remembers That you turned down the ORM in March, and what the migration would have cost.

excerpt five lines — four from this persona’s own sheet, which also holds its voice and where it came from, and one from the store only this persona builds.

What every persona can reach. Nothing on this shelf is locked to one seat: the critic and the chef call the same tools, read the same skill library and use the same catalog. What makes each one different is its sheet.

108 tools

Your assistant calls them; you never type one. The daemon lists 55 up front, and your assistant loads the others by name when it needs them.

  • ld_load_persona
  • ld_launch
  • ld_read
  • ld_write
  • ld_skill_get
  • ld_skill_add
  • ld_api
  • ld_api_call

20 skills ship

They sit in the shared library, open to every persona. A skill does nothing until you or your assistant call for it. Custom skills you write go in the same library, or into one persona’s own store.

  • Adversarial Reviewer
  • Analyze
  • Analyze Structure
  • Code Checker
  • Code-hunter
  • Constraints
  • Continuity
  • Create Persona
  • Create Plan
  • Devil’s Advocate
  • Launch
  • Launch Model
  • Poison
  • Review Plan
  • Serial Debate
  • serial-divergence-chain
  • Single source of truth
  • Skill Builder
  • Swarm Trace

141 third‑party APIs

Services in 30 categories ship in the catalog. Your assistant looks one up, makes the call, and the reply is kept in working memory for seven days. A service that needs a key calls with your key, set on your own computer.

  • SEC EDGAR
  • FRED
  • NIST NVD
  • VirusTotal
  • GitHub
  • PubMed
  • arXiv
  • Wikipedia
  • YouTube Data
  • Google Trends
  • Notion
  • Open Food Facts

tools: GET /api/status and tools/list · skills: the starter library in the installer · APIs: ld_api, two of the 141 marked dead in its own listing

Seven sections, not a prompt

An identity anchor, a description, a personality and a voice — then the two a prompt never gives you. A spine directive for what it must always do, and exclusion zones for what it must never do. A gas pedal and a guard rail, stored as fields, not pleaded for in a sentence that scrolls out of the window. A formative event closes it: where the persona learned to work this way.

It owns its own memory

Creating a persona writes it an identity record and a private store that only it builds, alongside the store you share. The security hunter’s notes and the engineer’s notes sit in different files, and nothing crosses between them unless you copy it across on purpose. They are SQLite files on your disk, and we hold no copy of them unless you choose to connect an assistant to our hosted endpoint.

Loaded, or launched

Load a persona and it becomes the assistant you are talking to — the sheet and its memories come into the session with it. Launch one and it is dispatched as its own agent, in its own session, working while you keep yours. That is how two personas review the same work at the same time and disagree about it.

Skills & Augments

Stop wasting time and tokens re‑explaining.
Start using memories and skills.

you type

it loads
   

An AI skill is stored with the condition it applies to, so you can ask for one by name or ask for whichever one fits. Augments equip into a skill’s sockets, and the version you build gets deployed under its own name. You wrote it once and never look it up again.

Works in the AI clients you already use

Easy API Access. Then just talk.

Write a memory in one client, read it back in another. Pick any two — it is the same store underneath.

write it in
read it back in
Claude Code · Monday

Remember we ship on Fridays, and I hate tickets with no owner.

Saved.

Cursor · Thursday

Draft the release checklist.

Friday ship date, and every item has an owner — I know you want that.

Nothing was copied between them.