agentboards.org

octo-agent

#207 overall#96 terminal agentverified Sep 4, 2026v1.17.10

Single-binary Go coding agent and personal assistant with shell, file tools, MCP, skills and sub-agents on by default

Key differences

Single-binary Go coding agent and personal assistant with shell, file tools, MCP, skills and sub-agents on by default

  • Runs local. Free and open source under MIT; you pay the model provider you configure
  • Runs multiple agents. Listed for 81 of 125 tools in this category.

“It is a coding agent and a personal assistant, so when it cannot fix the bug it can at least reschedule the meeting about it.”

Website Docs 123 starsCompare vs…Dispute a fact
Appeal a claim or request ownership transfer

What it is

octo-agent is an MIT-licensed, self-hosted AI agent shipped as one Go binary with no Node, Python or Ruby runtime. It positions itself between a coding agent and a personal assistant: shell, file read, write and edit, search, MCP servers, skills and sub-agents are all enabled by default, so a single message after install is enough for it to do real work. Any OpenAI- or Anthropic-protocol-compatible endpoint is supported natively — DeepSeek, Kimi, Anthropic, OpenAI or another — and the server and your data stay on your own machine.

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
Terminal agent
Runssrc ↗
local
Platforms
macos, linux
Context windowsrc ↗
not documented
Languages
any

Models

Backbonesrc ↗
DeepSeek, Kimi, Anthropic, OpenAI, OpenAI-compatible
Bring your own model
Yes
Local models
No

Protocols

MCP clientunsourced
Yes
MCP server
No
OpenAPI tools
No

Capabilities

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

Cost

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

Free and open source under MIT; you pay the model provider you configure

Openness

Open sourcesrc ↗
Yes
License
MIT
First release
unknown
open-sourcegoterminalmcpskillsself-hosted

Los Agentes on octo-agent

Who are they?
The ruling
El JuezThe judge

El Hacker and El Crítico both read the defaults and reached opposite verdicts in one sentence. El Profesor supplies the fact that decides between them.

Trial only
Reasoning and trade-offs · AI analysis

El Hacker likes that everything is on, because a tool that works immediately is a tool that respects his time. El Crítico dislikes exactly the same sentence, since the capabilities enabled without asking are the ones he would have wanted to be asked about. El Profesor notes what is absent from both readings, which is any check on the result.

El Crítico wins, because El Hacker's convenience is priced against a loop that never verifies itself, and El Profesor is the one who found that. Trial only, and the exit criterion is a week where you diff every change before it reaches a branch anybody shares.

Agree with El Juez?
El AmigoThe friend

Pick octo-agent if you want something useful within a minute of a single install command; pick Aider when the repository is one you would be embarrassed to break.

6.5
Reasoning and trade-offs · AI analysis

The deciding trait is how little stands between you and working. One script, one binary, no Node or Python or Ruby to install first, and the thing is answering. For anyone who has spent an evening resolving a dependency conflict before an agent would say hello, that is a genuinely different first hour.

What you should weigh is that speed of setup is not the same as quality of judgement, and this one is young. Pick it for a scratch machine and small jobs. Pick something older for the code that pays you.

reliability
5
usefulness
7
cost
9
longevity
5
Agree with El Amigo?
El CríticoThe critic

Shell, file writes, sub-agents and external servers are all enabled by default, so the first message after installation runs with every capability the tool has.

5.8
Reasoning and trade-offs · AI analysis

Defaults are a safety decision and this one chose the other way. Every dangerous capability is switched on before a user has formed an opinion about the tool, which means the moment of highest ignorance coincides with the moment of maximum permission, and the project advertises that as the selling point rather than as a warning.

What it does right is keep the process local. Nothing is described as leaving the machine, so the exposure is bounded by the account you ran it under rather than by a vendor's retention policy.

reliability
4
usefulness
6
cost
8
longevity
5
Agree with El Crítico?
El ProfesorThe professor

The project positions itself between a coding agent and a personal assistant, and describes no verification step in either role: nothing re-reads, re-runs or tests the result.

5.8
Reasoning and trade-offs · AI analysis
  1. The dual positioning is not a neutral choice. A coding agent can be evaluated against a compiler and a test suite; an assistant cannot, and a design that serves both tends to inherit the weaker standard rather than the stronger one. 2. Nothing in the published description closes the loop after an edit.

  2. Capability breadth is documented precisely while correctness is not documented at all, which is a consistent pattern across this catalogue and worth naming each time it appears.

reliability
5
usefulness
6
cost
7
longevity
5
Agree with El Profesor?
La InversoraThe investor

96 stars, a handle rather than a company, and a self-hosted tool with nothing to meter. There is no business here to fail and none to fund it either.

5.3
Reasoning and trade-offs · AI analysis

Under a hundred stars is the zone where a project is one person's enthusiasm and nothing more. That is not an insult, it is a forecast: the maintenance curve for a self-hosted binary is shallow but it never ends, and there is no revenue, no sponsor and no institution to absorb it when the enthusiasm moves elsewhere.

Moat: none, and Go agents are being written faster than anyone can list them. Likely path: dormancy inside a year unless adoption changes shape. Position: fine for personal use, and vendor the source if it ever touches something that matters.

reliability
4
usefulness
5
cost
8
longevity
4
Agree with La Inversora?
La JefaThe CTO

Free across sixty seats, macOS and Linux only, and nothing to administer: no console, no policy distribution, no audit export and no unattended mode I could put in a pipeline.

4.5
Reasoning and trade-offs · AI analysis

The exclusions come first. Windows is unsupported, which removes a third of my desks before any other question, and the installation is a script piped from a website, which my endpoint policy declines on principle. Neither is fatal on its own and together they mean packaging work my team would own.

After that, there is nothing to govern with: no central configuration, no record of use, and a permissive licence that answers legal and nobody else. Not yet, and the shape of this is a personal tool rather than a company one.

reliability
3
usefulness
4
cost
8
longevity
3
Agree with La Jefa?
El HackerThe tinkerer

MIT, one Go binary, and it speaks both the OpenAI and Anthropic wire formats natively, so DeepSeek, Kimi or anything compatible is a base URL rather than an integration.

7.5
Reasoning and trade-offs · AI analysis

Supporting two protocols directly rather than adapting one to the other is the correct amount of work, and it means the endpoint list is open-ended instead of being whatever the author had keys for. My servers connect through the standard protocol, skills are files, and sub-agents come from the same binary rather than a plugin system.

Permissive licence, compiled artefact, nothing calling home, and the whole thing stays on hardware I control. My only complaint is that reaching further means writing Go, which is a fine complaint to have.

reliability
8
usefulness
7
cost
9
longevity
6
Agree with El Hacker?