This paper presents a series of recent improvements made on the P300 speller paradigm in the context of the CoAdapt project. The flashing sequence is elicited by a new design called RIPRAND, in which the flashing rate of elements can be controlled independently of grid cardinality. Element-based evidence accumulation allows early-stopping of the flashes as soon as the symbol has been detected with confidence. No calibration session is nec-essary, thanks to a mixture-of-experts method which makes the initial predictions. When sufficient data can be buffered, subject-specific spatial and temporal filters are learned, with which the interface seamlessly makes its predictions, and the classifiers are adapted online. This paper, which presents results of three online sessions totalling 26 subjects, is the first to report online performance of a P300 speller with no calibration. 1 Material and Methods The P300 speller presented in this work was implemented in C++ with OpenViBE [7], and a dedicated stimulating software controlled the keyboard display. The software is opensource and part of OpenViBE release 0.18 We used a single Windows laptop to run all software components. The P300 speller keyboard was displayed on a separate LCD screen. A TMSi Refa8 amplifier, synchronized via hardware to the laptop, was used to record from 12 actively shielded electrodes. The visual stimulations consisted of briefly flashing "smiley" pictures. The P300 wave was detected via 3 channels of an xDAWN spatial filter [8], combined with a Regularized LDA classifier hereforth called RDA, which incorporates a regularisation of the common covariance matrix. The output of the classifier at each flashing time t is denote y(t). To save time, elements are always flashed in groups. Initial design of P300 speller groups involved rows and columns of a square matrix [2] or their randomizations [1]. The target element is then found at the intersection of the groups eliciting a P300 response. But repetitively flashing the same groups causes elements within the target groups to be wrongly selected, because of visual attention effects, and because of the contamination of all group elements by classification errors. Element-wise evidence accumulation avoids these two effects. A different random per-mutation can then be performed at each repetition of the flashes, effectively changing elements' group membership across repetitions. At each flash t, let the binary vector a(t) represent the set of n flashed elements within the grid of cardinality N . The score α(t) of each element (initialized to 0 at time 0) is updated with the following scheme, in which both target and non-target flashes contribute to the accumulation: α(t) = α(t − 1) + log