SmartDecision 0.4.0

Tue, 13 Oct 2026 00:00:00 GMT

A saved change that would break your apps no longer breaks them. If a decision has already answered API requests and you add an input, remove an output, change a field’s type, or add a new value to a text or true/false output, saving creates a new version. Requests without a SmartDecision-Version header keep getting the previous version until you retire it.
Response headers say which version answered and which version is the latest.

Without a license key, SmartDecision shows only the key screen.
When a license expires, everything keeps working for 14 days. From day 15 the workspace is read-only, and from day 31 the resolve API stops answering. Banners warn you before each stage, and a new key restores everything at once.
Trial keys can have shorter stages.

When the AI can’t draft a table, you now see why: the policy is too long for one decision, or the AI’s table had a problem it couldn’t fix. SmartDecision sends an invalid table back to the AI once to correct it before giving up.

The installer asks whether to use the built-in AI or your own AI server on your network (Ollama, vLLM or llama.cpp).

SmartDecision 0.3.0: First release

Mon, 05 Oct 2026 00:00:00 GMT

The first version of SmartDecision for customers. It installs on your own Linux server with one script and your license key.

Draft a decision table from a policy written in plain English. The built-in AI runs on your server. It starts loading when you open New decision and stops after 10 minutes without use, so it only takes memory while you draft.
Rules read as sentences: IF the conditions match, THEN these outputs. Open a rule to edit its AND and OR conditions, reorder rules by dragging or with Alt and the arrow keys, and add an optional Otherwise rule at the end. The table view is still there, and both views edit the same draft.
Inputs and outputs show above the rules. Create new ones from inside a rule. After a decision is saved, renaming a field changes its label only, so the API key your apps send stays the same until you change it on purpose.
Rules show their problems where they are: a rule that never runs, a value that can’t be compared, a missing output. Saving with problems asks you first and says what callers will get.
Your saved tests run against unsaved changes as you edit. The editor says “All 4 tests pass” or “1 of 4 tests would fail” and links to the rule that answered. Saving a draft that fails tests asks you first.

The first person to sign up becomes the admin. Everyone else joins with an invitation from Settings.
The activity log shows who changed what, with the before and after of each change.

install.sh downloads SmartDecision from get.smartdecision.io with your license key and starts it with Docker Compose. There is no registry to log into.
Choose the built-in AI, your own AI server, or no AI. The built-in model is Qwen3-4B-Instruct-2507: a 2.5 GB download that needs about 8 GB of RAM while drafting, and no GPU.
Enter your license key in Settings.

A true/false input set to any value no longer makes the resolve API answer with an error.
Ranges on inputs saved with the old text type compare as text instead of failing.