Everything it does, how each part works, and how it stacks up.
Waggle is an AI assistant that never forgets. Below: every major feature in plain language
(what it does and how it works), followed by an honest comparison with four other AI agents.
π‘ The one idea to hold onto: most AI tools forget you the moment a chat ends.
Waggle keeps a permanent memory for every project β and can even sit on top of the other AI tools you already use,
remembering across all of them.
1 Β· It remembers everything
The part nobody else does well. This is the core β and it's free forever.
π§ Persistent memory ("Hive-Mind")
Remembers every fact, decision, and preference β across days, weeks, and sessions.
How · Everything you tell it is saved as a "frame" in a local database. When you ask something, it finds the relevant memories by meaning and by keyword, links people & things into a knowledge graph, and keeps a live picture of who you are and what you're working on now.
ποΈWorkspaces
A separate, sealed space for each project. What's discussed in one never leaks into another.
How · Each workspace has its own private memory. Start from one of 15 ready-made templates (Sales, Legal, Engineering, Financeβ¦) that pre-load the right setup, and switch between workspaces freely.
π₯Harvest β import your past
Bring in your old ChatGPT, Claude, Gemini, PDFs, and web pages so it knows your history on day one.
How · Adapters read each export, strip duplicates, and an extraction step ("Cognify") pulls the meaningful facts straight into your memory.
2 Β· It does the actual work
Not just chat β it researches, writes, codes, and plans, and can run a whole team of agents.
π€Specialists / personas (22)
Switch the assistant's expertise on demand β researcher, writer, legal, finance, coder, analyst, and more.
How · 8 universal modes (including a read-only Planner, an adversarial Verifier, and a Coordinator that runs other agents) + 14 domain specialists matched to your workspace. Each has its own instructions, tools, and guardrails β e.g. the Verifier can only critique, never edit.
π οΈReal tools, not just talk
Web research, reading/writing files, running code, generating Word docs, building step-by-step plans.
How · Each specialist gets a curated toolkit (the Coder gets Git; the Writer gets document export) and a safety filter controls exactly what each one is allowed to touch.
πTeams of agents (WaggleDance)
For big jobs, it runs several agents in parallel and coordinates them.
How · A Coordinator breaks the job into pieces, spawns specialist workers, gathers their results, and a Verifier checks the output before it reaches you.
π§¬It gets better over time (Evolution)
The system learns to do recurring tasks better β not just the same thing every time.
How · It runs variations of a task, a judge scores them, an optimizer keeps what works, and improvements roll out behind safety gates.
3 Β· It connects, extends, and protects
Plug in your apps, sit on top of your other AI tools, and keep it all on your own machine.
πMarketplace (skills + connectors)
Add new abilities, and connect your apps β Gmail, Slack, Notion, GitHub, databases, and more.
How · Install skills and connectors from a catalog; connectors securely sign in to your services. This is the main paid upgrade.
πWorks WITH your other AI tools
Detects, launches, and connects other AI agents β Claude Code, Claude Desktop, Codex, Cursor, Hermes, OpenClaw β so their work feeds your memory too.
How · It installs reversible "hooks" into those tools; every session they run gets captured into your Hive-Mind. Waggle becomes the shared memory above all of them.
πPrivate & local
Runs as a desktop app; your memory lives on your own computer, not someone else's cloud.
How · A desktop binary with a local database; secrets in an encrypted vault; strict security β injection defense, sandboxing, no secret leakage.
πͺPlans & Enterprise (KVARK)
Free forever for memory; paid for skills/connectors/teams; enterprise runs sovereign on your own infrastructure.
How · Trial β Free β Pro ($19/mo) β Teams ($49/seat) β Enterprise = KVARK, the on-prem version where data never leaves your perimeter.
How a single request flows
01
You ask in plain words
β
02
It picks the right specialist
β
03
Recalls what it knows about your project
β
04
Does the work (research, docs, codeβ¦)
β
05
Saves what it learned to memory
β
06
Streams the answer + files back
βΊ Next time, step 03 already has everything step 05 saved β so it never asks you to repeat yourself.
How Waggle compares
βοΈ Played fair. Waggle actually integrates Hermes & OpenClaw (it can capture their sessions into memory), so these aren't pure rivals β they're tools Waggle can sit on top of. Researched mid-2026 from public sources.
Product
Runs on your own machine?
Remembers across sessions?
Team of agents?
Does real work (not just chat)?
Open source?
Best for
π Waggle OSEgzakta
β desktop app
ββ first-classper-project, free forever
β WaggleDance
β researchΒ·docsΒ·codeΒ·plans
βmemory engine OSS Β· app commercial
One memory across every project and every AI tool you use
HermesNous Research
β CLI/TUI/desktop
β self-improving
β subagents
β 40+ tools, shell
β MIT
Power users / devs wanting an always-on terminal agent
OpenClawP. Steinberger + community
β self-hosted
β local files
β channel-routed
β + drives other coding CLIs
β MIT
Devs wanting a private self-hosted agent + orchestration
Claude CoworkAnthropic
β desktop app
β starts freshmemory is bolt-on / beta
β auto sub-agents
β its whole pitch
β commercial
Non-dev knowledge workers delegating doc / data tasks
OdysseyNeptune Technologies
β cloud / browser
β cloud-hosted
β delegation
β workflows + data
β commercial
Small businesses automating back-office ops, no-code
π― The clean win
vs Claude Cowork (the highest-profile rival): it explicitly starts every task fresh. Waggle's persistent, per-project memory is the default β that's the sharpest contrast.
π€ Not just rivals
Hermes & OpenClaw are local-first and have their own memory β but Waggle integrates them. The pitch: "Waggle remembers across the agents you already use."
β The real edge
Not "local" or "memory" alone β it's the combination: one cross-tool memory + 22 specialists + importing your old chats + a marketplace, in one private desktop app.