After some digging and leafing through and reading: PicoLisp is brilliant. What it is: simply a primitive Lisp with a very compact language core and some very unusual design decisions.
First and foremost: no compiler, not even a bytecode compiler. PicoLisp's "virtual machine" is simply an efficient Lisp interpreter for Lisp SEXPRs (the linked lists you know from old Lisp interpreters).
Another point: only lists, symbols, and numbers as data types. No floating points (instead, simple ways to use scaled integers) and strings are either lists of characters (which are symbols again) or simply symbols.
Also rather unusual: constant dynamic binding of symbols, not lexical binding. Old Lisp systems also only had dynamic binding. Lexical binding makes compilers easier and eliminates some annoying sources of errors, but dynamic binding enables some very interesting programming techniques (which is why many Lisp systems have fluids - dynamically bound variables - such as the global parameters in Common Lisp).
Other peculiarities: designed from the outset for developing database-driven applications. It includes its own database that contains all the features of typical object-relational databases and its own GUI library that targets various HTML variants (pure HTML, HTML with Java, and HTML with JavaScript).
Adding Prolog as a query language to the database is only a consistent further development - the integration of Prolog and Lisp is not new in general. But to achieve all this in such a compact way is smart.
What excites me about it? Well, anyone who has read Paul Graham's articles about Arc - the 100-year language he is working on - will see astonishing parallels. PicoLisp also limits itself to the bare essentials, uses quite compact representations, and has only strongly reduced data types. The idea is the same for both: back to clean abstraction on known concepts, not to dilute the source representation through premature optimization (data type selection).
The result? A compact Lisp core with very interesting properties and a fairly simple way to put together web applications. But also suitable for typical scripting stories (once you've figured out how to get programs integrated, for example). Additionally, an interpreter with manageable source code (not 14,000 lines of C source!), which you can easily extend with the functions necessary for your own projects. Overall, this strongly reminds me of the old TCL versions, except that here a real Lisp is underneath.
If you want to read for yourself, here are some documentation links:
However, all the documentation is not quite complete - functions keep appearing in one of the documents that are missing in the reference, or functions are used that are not mentioned anywhere and whose operation is not entirely clear. But there is also the complete (and only sparsely) documented source code of all library modules for that. Yes, it sounds a bit like a text adventure, I know. On the other hand, the volume of source code is quite manageable.