SciCope 0.6.10

From 0.6.3: A new chat at the microscope starts knowing your lab: the notes the assistant keeps (your rules, the places on the plate, your programs), your experiments with their goals and results, and your routines — read once at the start, so you can run the same experiment again from a fresh chat. Later it opens them again only when you ask, so long chats stay light. Drop files or folders onto the assistant panel to hand them over: pictures to look at, tables and notes to read, a folder listed with a few of its pictures. Also: a Newport XPS stage shows each axis — positioner and driver port, speed, travel from–to and what the controller allows — in one table in the Stage settings; safety rules between stages ignore an axis that stays where it is (a plate that only rises keeps its rule); and publishing a new version no longer slows the download of updates.

From 0.6.2: An experiment the assistant runs is now one movie, from before its first move to its last answer — the plate before, every move and pick, the waits while it thinks, and what came after — instead of a few seconds after each step. Pictures a program takes go back to the assistant, each named with where the stage was, so it no longer drives back to look again. Counting in bright-field finds worms by how much darker they are than the lawn right around them and by their shape, so lawn wrinkles, specks, fibres and bead clusters are left out (the count says how many and why); it can look twice and mark which worms moved, and gives each worm’s stage position. Find the scale lets you choose which stage moves and on which two axes — X and Y for a Newport XPS stage, not its height — and keeps which way each axis moves the picture for tracking and positions. Also: the assistant can pause or stop an experiment when you ask; each step’s answer is a short line, so a long session uses much less of your AI subscription; a Newport XPS stage takes its travel from the controller (no more “parameter out of range” at the edge); changing the loaded program by step numbers keeps your steps; a wide assistant panel no longer squeezes the folder view.

From 0.6.1: The assistant uses less of your AI subscription in long runs. After it carries out a step, SciCope now tells it only what changed at the microscope — a stage position, the name of the new picture — instead of describing the whole microscope again every time, so each round carries about a third of the text it did. Experiments that run by themselves can use a smaller model of their own (Settings › Assistant › In experiments that run by themselves; the Experiment window shows which one), while the chat keeps yours; and one experiment task stops after 15 rounds instead of 30 when the assistant goes round in circles. Fixed: in an experiment task the assistant did not get the picture of its last step when that step made only one picture, so it counted worms without seeing them. It now always sees the picture.

From 0.6.0: Live animals are not lost. Press Pick the animal and click the worm right on the live picture. When the picker puts a worm down, SciCope compares a picture of the spot from just before with one from just after, finds the worm in a few hundredths of a second and follows it, so a fast worm cannot crawl away while the assistant thinks. Counting works for every kind of worm: GFP worms in fluorescence, all worms in bright-field, normal = all − GFP, and each count saves a picture with every worm circled and numbered, which the experiment report shows. Experiments are simpler to read and to change: each task is a card with one sentence for when it runs and its steps in order, a timeline shows every run, and a task can run several of your saved programs one after the other. An experiment survives a restart — SciCope opens again by itself, connects the cameras, stages and lights it had, and goes on — and it checks the picture before every task: a black or frozen picture skips that run with an alert, a blurry one is focused again. Pictures are saved as TIFF by default, with the name, number and format right in the Capture panel. In Analyze an imported sequence shows the picture alone (layers come on as you use them), the Tools and Export menus are in groups, and one Make a movie window covers every kind of movie.

From 0.5.26: Experiments that run for days without anyone at the microscope. Ask the assistant for one — say, three days of GFP and normal worms on one lawn — and it writes a schedule you can read: pictures, timelapses, your own programs and its own checks, each at its time. Start it, save it for later, change any task by hand or ask the assistant to redo the plan. The Experiment window lists what is running, what is saved and what has finished; SciCope keeps the computer awake, writes each measurement to a results table and builds an experiment report (PDF) with charts, a timeline and the pictures. Every experiment the assistant runs is recorded — a video when it is short, a timelapse when it is long — and gets a formatted report in its own folder, with the method taken from the settings that were really used. Safer runs: a program the assistant wants to run is walked through first, and the card shows where every stage will go and anything outside its travel before you allow it; safety rules between stages (for example, no X move while the picker is down); the lights switch off when a run is stopped; and a desktop alert when an unattended run fails, a camera stops or the disk is nearly full. Live tracking numbers every animal and keeps the number on it, and clicking a number follows that animal, with sharp marks that move smoothly over the live picture. Also: how long each light was on, saved with every picture; the Capture panel shows the next picture’s number; folders open in Finder or File Explorer from the app; and the picture viewer closes when you go back to the Microscope.

From 0.5.25: Simpler panels in the Microscope view — each panel keeps what you use while working (exposure, the jog pad, the light switches, Autofocus, the session and run) and everything you set once (rotation, scale, speeds and travel, which stage, relay or lens, line names, file names and format) moves to a Settings window opened from the ⚙ on each panel. A program’s light step lists the relay lines by the names you gave them, and a picture step lists the cameras; one light can hold several NI-DAQ relay boxes at once; a program that only switches lights, waits and takes pictures runs without a stage; and choosing another relay or stage from a list shows at once instead of after a click somewhere else.

From 0.5.24: a new SciCope logo and app icon — a microscope’s field of view with a worm curved into an S.

From 0.5.23: a simpler chat at the microscope — each command the assistant gives shows in plain words (“Look at the camera”, “Run the program”) while it runs and goes away when it has finished, without the commands’ values (Settings › Microscope › Show the assistant’s steps keeps them all); the model’s name fits in the chat (“Claude Code · Opus 5.5 · medium”); Timelapse and Run program have their own icons in the toolbar; a program step’s X, Y and Z boxes are wide enough to read; and the Timelapse window shows everything at once instead of scrolling sideways.

From 0.5.22: “Pick the worm” is faster and uses about a tenth of the AI subscription — the assistant runs the steps of your own loaded program by number, exactly as you wrote them (no retyping, no added wipes or holds), centres the worm and picks it up in one go (it goes on as soon as the worm is near the centre, and never picks when it did not get there), and moves the plate and puts the worm down in one go: three rounds instead of eight to twelve. Jobs at the microscope get short microscope-only instructions instead of the whole tool-building manual, and when several of your routines fit a request only the best one is used.

From 0.5.21: “pick the worm” checks its own work — after picking and after placing, the picker moves out of the camera’s view, the assistant looks at the plate, and it tries again (up to three times); the assistant always thinks at the effort you choose.

From 0.5.20: routines — fixed step lists the assistant follows for jobs you repeat, so it does them the same way every time and quickly. “Pick the worm” is built in (find the worm, centre it, run your pick-and-place program, check the drop); the assistant saves a procedure that worked as a new routine when you ask, later chats use it, and Settings › Microscope › Routines shows, edits and deletes them. It never reads your other chats unless you ask.

From 0.5.19: a fully automatic mode for runs nobody watches — with Fully automatic switched on in Settings › Microscope, the assistant does not wait for any Allow, homing and device changes included (the red STOP button and Escape still stop everything), and the assistant is told which mode is on.

From 0.5.18: the assistant uses far less of your Claude subscription — the Claude Code app is started with only what SciCope needs (its own coding instructions, extra tools and plug-ins left out), the pictures go inside the message instead of being opened one by one, and a conversation that has grown large continues in a fresh session that carries the text over without the old pictures.

From 0.5.17: the assistant runs your experiments itself — it sees the result of every command it gives, with its pictures (a look at the live camera, what the tracker sees, the animal it locked on), it knows every stage of the microscope by name with its axes, travel and speed, and your loaded program, and it works with that program instead of writing a tool. With Let the assistant run experiments by itself switched on it goes ahead on its own and still asks you before homing, resets and changing devices. The assistant’s instructions and knowledge are sent once per conversation, not with every message, which uses much less of your AI subscription; the Claude Code app has an effort setting (Low to Max). In the Microscope view: a Run program button next to Track, and the toolbar’s Autofocus works with any stage axis, chosen from the ▾ beside it. A Newport XPS stage shows its own top speed (300 mm/min for the LTA-HS) beside each speed box, in one unit. The chat box no longer hides the first letters you type.

From 0.5.16: a program step can move several axes at once — go to X 5 and Y 5 together instead of one after the other: the step has a box for each axis of its stage, and the axes you fill move together (one straight move on an Arduino board; on a Newport XPS stage the X/Y group and the Z group now move at the same time); and the numbers typed in the stage’s Go to boxes stay there after the move instead of being cleared.

From 0.5.15: loop programs work with several stages — each step that moves something says which of the microscope’s stages it moves, a scan picks its stage and its two axes, Y steps are allowed (they were refused with “Move steps need axis X or Z”), and an axis the stage does not have is refused with a sentence; clicking or painting the worm takes in its whole body, its paler parts too; the animal followed is shown in green in the tracking window; the “Find the animal” page is organized in groups, with a number box beside every slider; and the autofocus lands on the sharpest point — its coarse sweep sees which way focus lies even when its steps are much larger than the depth of focus, the last step is fitted between the measured positions, and a stage axis always approaches from below so play in its screw does not shift the focus.

From 0.5.14: live tracking that holds on — after each correction the tracker looks where the move has taken the animal (a correction larger than “max jump” used to lose it for good), it waits for a picture taken after the move, an animal that touches its trail, breaks into pieces or reaches the border is still followed, a lost animal is found again, and you can follow a whole swarm by the centre of all its pixels; the “Find the animal” page is the Detect panel of Analyze, live, with the same sliders and settings; the tracking window says Stop tracking while it runs; autofocus can use any stage axis, such as the Z of a Newport XPS stage, and refuses an axis that slides the picture instead of focusing it, putting the stage back; and every stage’s jog step can be one of your own.

From 0.5.13: the live worm tracker finds animals with the same detection as Analyze, so “Faintest animal kept”, joining the pieces of one animal and splitting touching animals now work while tracking, and one click takes the Detect settings you tuned in Analyze; “Identify the stage” now shows the mm per pixel it measured (Save no longer puts the old value back), and it works from the picture itself when no animal is found; the Photos view shows the folders inside a folder, such as “SciCope analysis”, and opens them; and the fast timelapse keeps the interval you typed instead of going back to 0 s.

From 0.5.12: Newport XPS controllers with several stages — the XYZ stage and the YZ stage on one controller, each set up in the app with its own axes and driver ports, and a port can be changed and written to the controller from XPS setup; a microscope can drive more than one stage, each with its own card; STOP aborts a stage’s move without switching it off; and the tracker follows an animal with the stage and the two axes you choose, identified with a small test move.

From 0.5.11: the Program panel — add a grid or spiral scan straight from “Add step”, drag steps to reorder them, pick several at once, and watch a clock that counts down every wait (it stands still while paused); a larger worm-tracking window with its settings on three short pages; chats with the Claude Code and Codex apps continue their own session instead of re-sending the whole conversation each time, which uses far less of your subscription; and the newest AI models in the lists (Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5, GPT-6, Gemini 4).

From 0.5.10: the live picture of an Andor camera is shown the way Andor Solis shows it — from the darkest to the brightest pixel, so nothing clips — instead of a strong stretch that turned dark parts black and bright edges white; a Display choice for every 16-bit camera (Min–max, Stretch, the full 16-bit range, or two counts you lock), a histogram and the picture’s 16-bit counts beside it, and saved JPEGs that look like the screen. Clicking a button in a panel no longer sends the panel back to the top.

From 0.5.9: Andor EMCCD cameras without Andor’s own software — the shutter opens by itself, every setting (cooling, EM gain, readout, vertical shift, baseline, triggers, frame transfer, binning, area, crop mode) in the Camera panel and one window; the Andor and Thorlabs cameras plugged in are listed the moment the app opens and opened only when you choose one. From the versions before: an Arduino stage board drives X, Y and an up/down Z, each switched on or off; Controller settings and Devices in pages; the Photos view previews every file in a folder (16-bit TIFFs with automatic contrast, ImageJ stacks, pictures of hundreds of megapixels, movies, tables, Word, PowerPoint, PDF), Settings in pages with a search box; several microscopes in one window, a tab for each, every tab an app of its own; the ZEISS controls spread over Stage, Light, Focus and Objective with an autofocus for every focus drive; pictures saved in Documents/SciCope unless you choose another place; and in Analyze, layers like a photo editor and a plate map.