Updates while it runs
An orchid opens new flowers while it keeps growing. Nodecode makes its updates on the fly and the program evolves naturally.
A living agent.
It becomes yours and evolves over time.
Subscribe to our finetuned uncensored models · Pricing
$ curl -fsSL https://nodecode.ai/install.sh | sh- Copy
- Paste in a terminal
- Run
nodecode
Everything is a cell
A cell is like a folder that teaches Nodecode something new: a provider login, a tool, a skill, a GitHub repo, a MCP server, you name it.
/evict, no more /compact
Compaction rewrites older turns into a summary and loses the detail. Eviction moves them to the log whole, and every word can be read back. Accuracy holds: a new paradigm for long sessions.
“Its programs are in the same format as its data.”John McCarthy · 1984
Lisp
Lisp evolves its own language: code is data, so a program can rewrite itself while it runs. Nodecode redefines itself naturally.
Cells, the building blocks of the organism.
Cells hub
A cell is a folder that teaches Nodecode something new: a chat room it can live in, a provider login, a tool, a skill, a GitHub repo, you name it.
From Nodecode 37Show all
nodecode-aiProvidersnc://clineCline's API: the client header its free models need, and its model feedPinned 25 Sepv0.1.0nodecode-aiProvidersnc://codex-authChatGPT Codex OAuth credentials for openai-family lanesPinned 25 Sepv0.1.0nodecode-aiClientsnc://channel-slackTalk to nodecode through a Slack appPinned 25 Sepv0.1.0
Ships with Nodecode 14
1(nle:define-cell weather2 (:section ("weather")3 ("units" :choice4 :options '("c" "f"))5 ("cities" :list))6 (:tool "weather" +schema+ 'look-up)7 (:command "weather" 'run-slash)8 (:hook :frame #'observe))
One form declares a cell: its settings, tools, commands and hooks.
nodecode add❯ nodecode add nc://cline
cline: Cline's API: the client header its free models need, and its model feed
from https://github.com/nodecode-ai/addons (nodecode-cline/)
commit c1e7763424a6a26723f0aae2053b6d79644f127d
It runs inside Nodecode with full access to your files and keys.
Install cline? [y/N] y
cline is on
❯
Add a cell.
- Your repogithub.com/you/weather
- Read at a commitweather.asd and its README
- Listednc://weather
Upload your own cell. Anyone can use it.
Subscribe, our finetuned, uncensored models are completely private.
Founders
$30/moexcluding taxes
About 12 full workdays of an agent coding nonstop. A coding agent makes about 200 requests an hour in steady work, measured on 470 of our own sessions, so 20,000 is about 100 hours. a month
4 requests at the same time
Subscribe
Cancel any time.
Founders Plus
$90/moexcluding taxes
About 50 full workdays of an agent coding nonstop. A coding agent makes about 200 requests an hour in steady work, measured on 470 of our own sessions, so 80,000 is about 400 hours. a month
8 requests at the same time
25% less per request
Subscribe
Cancel any time.
Protecting our customers
- ZDR
- Zero data retention. Your prompts and answers are private and never stored on our servers.
- E2EE
- End-to-end encryption. Your requests are encrypted from your agent to the machine that runs the model.
- GDPR
- Data protection. Your requests are processed in Europe, only to answer you, under European privacy law.
0
prompts or answers kept, fully private.
On both plans
GLM 5.3 FlashCyber security finetune
Qwen 4 27BResearch finetune
DJevDecision model
Qwen-Image 2.1 TurboImage model
2 months free on yearly plan$300 a year for Founders instead of $360.
Research. Grounded on research via Foresight Labs.
Claude Code Claude Opus 5.5 · 100Nodecode Claude Opus 5.5 · 100Codex CLI GPT-6.1 Sol · 99.9Nodecode GPT-6.1 Sol · 99.5Human baseline 95.4%0%25%50%75%100%080M160M240M320M400MAll tokens, games in a row · linear
54% fewer tokens: 181M against 393M, in 2,333 against 2,241 model calls. 41% less in costs: $124 against $210 at public list prices. The same model, Claude Opus 5.5, on all 25 games, each game's best run, both scoring 100. than Claude Code
66% fewer tokens: 58M against 172M, in 2,822 against 2,298 model calls. 24% less in costs: $33 against $43 at public list prices. The same model, GPT-6.1 Sol, on the same 25 games, each game's best run. Codex scores a little higher, 99.9 to 99.5, with fewer actions per level. Codex had more than one run at 8 games and Nodecode at 3, and 20 of Codex's games ran under a 1,000-action cap, every other game under 1,500. than Codex on GPT-6.1 Sol, on novel interactive reasoning tasks such as ARC-AGI-3.