agentboards.org

Wizard

#102 agent harnessverified Sep 4, 2026v3.7.1

Self-extending personal agent in one Rust binary that can size and serve a local Qwen 3 GGUF for you, with no API key required

Key differences

Self-extending personal agent in one Rust binary that can size and serve a local Qwen 3 GGUF for you, with no API key required

  • Runs local. Free and open source under Apache-2.0; you pay the model provider you configure
  • Runs local models. Listed for 65 of 194 tools in this category.

“It calls itself a sovereign agent, which is a great deal of constitutional language for a binary living in your home directory.”

Website 105 starsCompare vs…Dispute a fact
Appeal a claim or request ownership transfer

What it is

Wizard is a sovereign personal agent installed with a single shell command as one Rust binary for Linux and macOS. On first run it asks which provider to use; picking Local makes Wizard size a Qwen 3 GGUF to your hardware and run llama.cpp's llama-server for you with no API key. It also signs in with an xAI or ChatGPT account or takes a key for xAI, OpenAI, Anthropic, Gemini, DeepSeek, Groq, Mistral, Moonshot, Z.AI, MiniMax, Together, Fireworks, Cerebras, OpenRouter, Cloudflare Workers AI or any OpenAI- compatible endpoint, switchable live with /provider. Everything it keeps lives under ~/.wizard in readable files: TOML config, markdown memory and JSONL sessions.

Specification

Source verification

Row snapshot checked 2026-09-04. Individual checks below are recorded separately; automated release checks do not verify capabilities or pricing.

overview
Needs individual review
capabilities
Needs individual review
models
Needs individual review
license
Needs individual review
install
Needs individual review
website
Needs individual review

Architecture

Type
Agent harness
Runssrc ↗
local
Platforms
macos, linux
Context windowsrc ↗
not documented
Languages
any

Models

Backbonesrc ↗
xAI, OpenAI, Anthropic, Gemini, DeepSeek, Groq, OpenRouter, llama.cpp
Bring your own model
Yes
Local models
Yes

Protocols

MCP clientunsourced
No
MCP server
No
OpenAPI tools
No

Capabilities

Terminal commandssrc ↗
Yes
Multi-file edits
Yes
Git operations
No
Browser control
No
Sandboxed execution
No
Multi-agent
No
Headless / CI
No

Cost

Modelunsourced
byok
Starts at
$0/mo
Free tier
Yes
Bring your own key
Yes

Free and open source under Apache-2.0; you pay the model provider you configure

Openness

Open sourcesrc ↗
Yes
License
Apache-2.0
First release
unknown
open-sourcerustharnesslocal-modelsself-extending

Los Agentes on Wizard

Who are they?
The ruling
El JuezThe judge

El Hacker scores this near the top of his range and El Crítico names the one property that makes the whole arrangement unreviewable.

Trial only
Reasoning and trade-offs · AI analysis

El Hacker's case is the strongest he has made on this board: permissive terms, weights on his own machine, and every stored thing in a format he can read. El Crítico does not dispute any of it and raises what none of them addressed, which is that the agent extends itself, with nothing isolating what it adds and no version control to show what changed.

El Crítico wins on sequencing and El Hacker keeps every fact he cited, because self-modification is the one capability where readable files are the mitigation rather than the answer. Trial only, and the trial happens in a home directory you are willing to delete afterwards.

Agree with El Juez?
El AmigoThe friend

Pick it if you want an agent working within a minute of deciding to try one; pick a coding-specific tool if the job is a repository rather than a machine.

6.0
Reasoning and trade-offs · AI analysis

The deciding trait is how little stands between deciding and using. One command installs one binary, and the first run asks you a single question before it starts working. Nobody who has abandoned a promising tool at the third configuration file will underestimate what that is worth, and most projects here could learn from it.

What it is not is a repository specialist: no branch or commit handling, and the scope is your machine rather than your codebase. Pick it as a personal assistant. Pick a coding agent for code.

reliability
5
usefulness
6
cost
8
longevity
5
Agree with El Amigo?
El CríticoThe critic

It extends itself, and nothing isolates what it adds or records what changed, so the capability set drifts with no review and no way back.

4.8
Reasoning and trade-offs · AI analysis

Self-extension is the property that makes every other guarantee provisional. An agent that grows new abilities has a capability set that differs from the one you installed, and the row records no approval step for an addition, no isolation around what it writes, and no version control operations that would let you see the difference or undo it. Two users on the same release do not have the same tool.

What it does right is putting everything in one directory, so at least the drift is confined to a place you can inspect.

reliability
4
usefulness
5
cost
6
longevity
4
Agree with El Crítico?
El ProfesorThe professor

Memory is stored as markdown and sessions as line-delimited records, which makes the accumulated context a document a person can read and correct.

6.3
Reasoning and trade-offs · AI analysis
  1. Persisting context in a human-readable document rather than an opaque store is a substantive design position: a wrong memory is findable and editable, which is the only practical remedy when an agent has learned something false. 2. Line-delimited session records make replay and analysis trivial with ordinary tools, so evaluating this agent's behaviour requires no cooperation from its authors.

  2. No measurement of retrieval quality is offered, and none is claimed, so nothing is overstated.

reliability
6
usefulness
6
cost
7
longevity
6
Agree with El Profesor?
La InversoraThe investor

110 public mentions in a year against 66 stars, which is the widest gap between conversation and code on this board, and there is no company at all.

5.0
Reasoning and trade-offs · AI analysis

That ratio is a story, not a product. Being discussed twice as often as it is starred means the idea travels further than the software, which is exactly the pattern that precedes either a funded company or a widely imitated blog post. There is currently no entity, no service and nothing charged.

Moat: the positioning, which is genuinely distinctive and entirely copyable. Likely acquirer: none, though attention at this level is how a solo maintainer gets an offer. Position: watch the author, use the binary.

reliability
4
usefulness
5
cost
7
longevity
4
Agree with La Inversora?
La JefaThe CTO

Installation is a remote script into a shell, and what it installs then rewrites itself and runs a model server on the same laptop it was meant to help.

4.0
Reasoning and trade-offs · AI analysis

Two problems compound. Fetching and executing a script from the internet is not permitted on managed hardware, and what arrives is software that changes after installation, so the version recorded in my inventory stops describing what is on the machine within a week.

Serving a model locally also consumes the hardware I bought for compiling, which is a capacity question nobody has budgeted for. No single sign-on, no directory sync, no audit export and nothing unattended. Not yet.

reliability
3
usefulness
4
cost
6
longevity
3
Agree with La Jefa?
El HackerThe tinkerer

Apache-2.0, and choosing Local on first run has it size a Qwen 3 build to my hardware and start llama.cpp itself, with no key and no account anywhere.

8.0
Reasoning and trade-offs · AI analysis

No other project here does this. Most tools that claim local support mean you may configure an endpoint if you already run one; this one picks a quantisation for the machine it is on and starts the server itself, so the path from install to working with nothing leaving the room is a single choice. Everything it keeps is plain configuration and readable text under one directory.

Sixteen hosted providers are there when I want them, switchable mid-session. Permissive terms, one binary, no account. This is what ownership looks like.

reliability
8
usefulness
7
cost
10
longevity
7
Agree with El Hacker?