KITvsKon
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
KIT
Mark3 Labs · Terminal agent
#34OSSMCP
Panel
6.83 spec wins
- Reliability
- 6.3
- Usefulness
- 6.7
- Cost
- 8.3
- Longevity
- 5.8
“KIT stands for Knowledge Inference Tool, which is what you name something when the good acronym was taken.”
Kon
0xku · Terminal agent
#83OSS
Panel
6.82 spec wins
- Reliability
- 6.3
- Usefulness
- 6.3
- Cost
- 9.0
- Longevity
- 5.3
“It ships themes, in a tool whose entire argument is that it refuses to waste tokens on anything decorative.”
Spec by spec
| Spec | KIT | Kon |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local |
| Platforms | macos, linux | macos, linux |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | No |
| Browser control | No | YesOptional web_search and web_fetch tools are enabled with --extra-tools; there is no driven browser. |
| Sandboxed execution | No | No |
| Multi-agent orchestration | Yes | No |
| Headless / CI mode | Yes | Yes"kon -p" runs one prompt non-interactively with tools auto-approved and returns documented exit codes. |
| Models | ||
| Backbone | Anthropic, OpenAI, Google Gemini, Ollama, Azure OpenAI, AWS Bedrock, OpenRouter | OpenAI, OpenAI Codex, GitHub Copilot, Azure AI Foundry, DeepSeek, xAI, Zhipu |
| Bring your own model | Yes | Yes |
| Local models | Yes | YesAny OpenAI- or Anthropic-compatible local endpoint works, with auth checks relaxable through --openai-compat-auth none. |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | $0/mo | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 138 | 354 |
Which one would each critic pick
| Critic | KIT | Kon | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.0 | 7.3 | Kon — Pick it if you resume yesterday's work more often than you start something new; pick a heavier agent when you want it to hold more of the project. |
| El Crítico | 6.3 | 6.0 | KIT — Sub-agents are markdown presets and their only boundary is a per-agent tool allowlist, which is a text file standing between a generated instruction and a built-in shell. |
| El Profesor | 7.0 | 7.8 | Kon — The core prompt is stated at under 270 tokens and the whole fixed harness at about 1,000, which is an accounting almost nobody in this category publishes. |
| La Inversora | 6.3 | 6.0 | KIT — 128 stars, permissive licence, no company and nothing to charge for, from a lab whose reputation was built giving away protocol plumbing rather than selling products. |
| La Jefa | 6.3 | 5.8 | KIT — It scaffolds a GitHub Actions workflow that reviews pull requests on a comment, and it has a non-interactive JSON mode, which is the first thing here I can put on a dashboard. |
| El Hacker | 8.0 | 7.8 | KIT — MIT, and custom tools, commands and widgets are written in Go and compiled in. It runs as an ACP agent over stdio, and Ollama is a first-class provider. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.