What's new in AgenticFM

v1.10.0 · October 14, 2026

• Find in the chat panel: Press Cmd+F to search the conversation, then Cmd+G and Shift+Cmd+G to jump between matches.
• New scripts land where you want them: An agent can create a script inside a folder or right after another script, and the card tells you which folder it went into.
• Layout mode stays yours: Choose Leave, Ask or Stay in the panel to decide whether an agent may take a window out of Layout mode before it runs a script.
• Run more than one FileMaker: Each FileMaker version gets its own panel title, toolbar icon and color, its own log and its own remote port, and settings stay safe when two copies of FileMaker save at once.
• Scripts that start when asked: The plug-in now knows when a script it queued has actually started, nudges FileMaker when a queued script is stuck waiting, and fails fast when a script halts before it answers instead of waiting out a timeout.
• Bridge upgrades that respect your work: Upgrading the bridge script now steps aside when you have a script paused in the Debugger or held by a dialog, names that script on the Update banner and keeps Retry ready. A stale bridge is replaced instead of refused.
• Script edits that round-trip cleanly: Comments keep their spacing and line breaks, quotes and semicolons inside text no longer swallow a step, missing tables and fields read back as FileMaker shows them, and Save Records, Export Records, Save a Copy as, Insert from Device and Perform AppleScript now read and write exactly the way FileMaker draws them.
• Faster, safer work on long scripts: Reading, inserting and patching long scripts no longer freezes FileMaker, adding steps to the top of a long script no longer crashes it, and each edit is reviewed on exactly the lines it wrote.
• Links and navigation land on the right script: Jumping to a script works after it was renamed, picks the right one when names differ only by case, and opens the right file when a link points to another file.
• A tougher security gate: The calculation gate now reads text, comments and punctuation exactly as FileMaker does, including fullwidth characters, decides on the real program behind a shell command, and blocks more risky MBS functions. Edits you make to your security lists take effect without a relaunch, and a newer shipped list is offered when yours was customized.
• Clearer chat cards and notices: Tool cards finish in green or red with a title that says how the run ended, notices can come and go on a timer you set, notices asking you to act stand out, and a stuck request has a Stop button.
• SQL in the file you meant: SQL queries, data changes and schema changes pinned to a file now run in that file, and creating a table names the File Options setting when FileMaker gives it no fields.
• Steadier day to day: The chat panel survives a server restart, a click on the panel is no longer swallowed after a FileMaker window takes focus, a mistyped display setting no longer blanks the panel, and the license banner always reflects your current license.

v1.9.1 · October 1, 2026

• Stray plug-in copies get caught: If an older or renamed AgenticFM sits next to the current one, the plug-in finds it by its real identity, moves it out of the way and tells you, so FileMaker always loads the version you meant.
• Your settings are safe from a bad file: If a settings file ever can't be read, the plug-in sets it aside instead of wiping it back to defaults, and your chosen AI provider is never quietly switched back to Claude.
• Models you can't use yet stay in view: If your AI command-line tool is too old for a model, or your account doesn't include it, the model stays in the picker, dimmed with the reason, and clicking it shows how to unlock it.
• Know when your AI tool has an update: When a newer version of your active provider's command-line tool is out, the chat tells you once and shows the command to update it. Rediscover now finishes in a few seconds instead of half a minute, and switching providers skips the confirm when there's no conversation to lose.
• Chat cards tell the whole story: Tool cards now show their results, a Run Script card follows the script from queued to finished, and when a SQL query, clipboard edit or script edit is refused, both you and an agent see why, so an agent can fix the request and try again instead of claiming it worked.
• Outside agents can work with your data: An agent connecting from another machine can now make SQL data and schema changes through the plug-in, not just the built-in chat, still subject to your modification policy settings.
• Inspector help on FileMaker 2026: Ask an agent where an Inspector setting lives, or to change one, and it can do it on FileMaker 2026.
• More reliable Script Workspace work: The plug-in now spots unsaved scripts hiding behind the Show more scripts button and in separate Edit Script windows, reads script names correctly even with extra columns from other plug-ins, and refuses an edit with a bad insert position instead of dropping steps at the end of the script.
• Works however you've set up FileMaker: The plug-in no longer kicks a window you're arranging in Layout mode back to Browse mode, it finds your active script even with the Scripts pane hidden, and links that open Manage Database work from the Script Workspace.
• Links and bridge calls that land: Deep links in chat bring FileMaker forward and report when they finish, scripts are recognized as done the moment FileMaker's script stack empties, and the plug-in waits instead of starting a background check while you have unsaved script changes, and a script that returns a plain value or list to an agent now keeps its result.
• Bridge upgrades don't stall: Upgrading the bridge script no longer gets stuck when FileMaker is in front but sitting idle, with Keep FileMaker Foregrounded left on its default of Ask.
• Chat speaks up when something doesn't land: A deep link in chat that can't open now shows a message instead of doing nothing, and when an agent asks for both your open script and the clipboard in one answer, the chat tells you which one wasn't added.
• Script chips stay with their file: Turning on a script for one file no longer turns on a script with the same name in another open file.
• Clipboard sizes you can trust: The clipboard header and its chip above the message box now show the size actually sent to the model. Hover to compare it with the raw clipboard size.
• Notices that notice you: When a long script only goes out in part, the chat says so with a link to the Max Steps in Prompt setting, and any notice with an Open setting button goes away once you change that setting.
• Smarter multi-file solutions: When two open files look like the same solution, AgenticFM asks whether one is a renamed copy, files with nothing in them show up as missing, and removing a file from a solution frees its disk space.
• Solution Intelligence answers about the right file: With several solutions analyzed, asking about any one of them now uses that solution's own data instead of whichever you analyzed last, and script and layout counts match what you see in FileMaker.
• Calculation checks catch quoted booleans: A drafted calc that compares against "true" in quotes, which never matches in FileMaker, now comes with a warning so you can fix it.
• Private by default in your logs: API keys, tokens and passwords are scrubbed from every log line, so a log you share with support stays clean.
• Steadier under load and at shutdown: Fixed several crashes and hangs when quitting FileMaker, and the plug-in keeps capacity in reserve so it stays responsive even when it's busy.

v1.9.0 · September 22, 2026

• Ask an AI to write your calculation, right in the Calculation dialog: There's a new Send Prompt button inside FileMaker's calculation editor. Describe what you want in plain language and an agent writes the calc for you, checking it and repairing it in a loop so only a calculation that actually validates ever lands in the box. It knows the functions your FileMaker version supports and the fields and table occurrence you're editing in, so what it writes fits your solution. Enable or disable the feature within Settings, and change the starting prompt prefix from PROMPT: to whatever you desire.
• Decide what happens when a calculation is ready: A new setting lets you choose whether a finished calculation brings its dialog straight to the front, quietly notifies you, or waits for you to come back to it, so a calc that finishes while you're working somewhere else lands where you expect instead of stealing your focus.
• An agent can build tables and fields for you: AgenticFM can now create new tables and fields in your solution with full FileMaker fidelity, including auto-enter and validation options, and it fails loudly with a clear reason when a name can't be created or an option can't be honored, instead of quietly leaving a dialog parked or dropping a setting.
• Your secrets stay out of the conversation: Before your solution's context is sent to an AI, the plug-in now scrubs out things that look like passwords, API keys, and other credentials, and shows you a plain notice of what it redacted. A new Privacy setting lets you turn scrubbing off if you need to, with an up-front warning about what that exposes.
• FileMaker 2025 and 2026's newest steps are fully supported: An agent can now read and write the latest AI-era script steps, including Configure AI Account, Insert Embedding, Perform Semantic Find, Insert Image Captions, Generate Response from Model, and Fine-Tune Model, plus NFC Reading, Configure Local Notification, and Trigger Claris Connect Flow, each round-tripping back into FileMaker exactly as you'd build it by hand.
• Snappier Script Workspace navigation: Jumps and focus changes in the Script Workspace now wait on FileMaker's real state instead of fixed delays, so navigating to a script or window lands quicker and more reliably.
• A round of crash and hang fixes: We hardened the paths that read scripts, calculations, layout objects, and the Data Viewer out of FileMaker's interface, fixing several crashes and hangs that could hit during those reads, and a button press that raised a modal now fails cleanly instead of deadlocking.
• Fresh model lineup and clearer discovery: Refreshed the Claude, Codex, and Cursor model rosters with smarter defaults for FileMaker work, and an outside agent connecting to the plug-in now sees its version and build up front and gets a helpful JSON reply when it calls an endpoint that doesn't exist.

v1.8.1 · September 10, 2026

• Write and run real tests against your scripts: AgenticFM has a new script-testing framework, so an agent can write a test suite for your scripts, run every case, and grade the results. Cases can set up and tear down their own data (including truncating a table to start clean), read fields, globals, the found set, layout, and mode out of band to check what a script actually did, and time each script so you can see what's slow. It runs across multiple files, and any test that touches your data stops at a consent card first so nothing gets modified without your say-so.
• Update the bridge across every open file at once: When more than one file needs the AgenticFM bridge refreshed, the chat panel now shows a checklist of all of them with a single "Update all files" button, instead of walking them one at a time. The upgrade itself got a faster path that skips Open Quickly, and the paste-and-restore under load is hardened so a bridge update can't quietly land empty or half-applied.
• Snappier replies, especially for solution-level questions: Ask a broad question about your whole solution and an agent now answers from a compact Solution Digest instead of re-reading everything, and the plug-in trims the amount of data it sends each turn. Responses come back noticeably quicker, and the digest shows up as its own inspector card you can open.
• Effort matched to the question: The plug-in now reads how much thinking a turn actually needs and tells command-line providers to match it, so quick asks stay quick and hard ones still get the deep reasoning.
• Script edits are even harder to wreck: An edit that would empty your whole script, or one that would paste the entire script back on top of itself and duplicate every step, is now caught and refused instead of committed. Deletes are double-checked to confirm they actually landed.
• More script steps round-trip faithfully: We kept comparing steps against a live copy of FileMaker and fixed more stragglers, including a Perform Script that calls into another file, Sort Records keeping its exact sort order, Insert Text with a semicolon in it, and text holding smart quotes that used to get mangled. An agent can also insert just one half of a block pair (an If without its End If, say) when that's what a surgical edit needs.
• A guard against duplicated solution files: If you've made a copy of a file the old-fashioned way so two files share the same internal identity, analysis used to get them confused. AgenticFM now spots the collision up front and tells you how to give the copy a fresh identity instead of quietly mixing them up.
• Clearer sign-in trouble with Claude Code: An expired Claude Code login is now recognized as an auth problem rather than a generic failure, and when the plug-in can quietly refresh it for you the banner says so instead of alarming you.
• Straight answers from data operations: Running a query with parameters now binds them correctly, and a data write (insert, update, delete) reports how many rows it actually changed so you're not guessing whether it took.
• Point saveAsXml at a normal file path: Handing the file-save a plain Mac path (the kind you'd copy from Finder) now works, instead of requiring FileMaker's own path format.
• Save tells the truth when FileMaker says no: If FileMaker refuses to save a script because it's invalid, that comes back as a clear failure now, not a misleading "saved fine."
• Navigation you can watch land: Jumps and window raises now settle and confirm before reporting success, so a navigation that worked actually looks like it worked.
• Crash fix around the Script Debugger: Reading the Data Viewer right after halting a script, while the Script Debugger window was still open, could crash FileMaker. That read now stands down until the debugger is closed instead of taking FileMaker down with it.
• New sign-in help: Added a "Signing in to Claude Code" walkthrough to the built-in help, with screenshots, so first-time setup is a lot less guesswork.

v1.8.0 · August 26, 2026

• Attach files right in the chat: There's a new paperclip in the chat prompt area, and you can drag and drop too, so you can hand an agent images, PDFs, and documents as part of the conversation. Each attachment shows up as a chip you can include or drop from your next turn, and a big attachment gets a size check and a delivery choice so a giant file can't quietly blow up a request.
• Agents can read what's inside your attached files: Attached documents and PDFs get their text pulled out for you, including scanned pages read via OCR, plus thumbnails, so an agent can work with the contents of a file you drop in rather than just its name.
• Work with FileMaker files as first-class attachments: Point AgenticFM at an .fmp12 file and it treats it as its own kind of attachment, referencing it in place instead of copying it around. It notices when that file changes on disk and flags the drift, and it can reveal the file in the Finder rather than opening it out from under you.
• Open protected FileMaker files: For an .fmp12 file that needs an account login or is encrypted, a new dialog collects the credentials, tucks them safely into your login Keychain, and reuses them the next time you reach for that file. Saved credentials you haven't touched in a long while get swept out automatically, and you decide how long is too long.
• FileMaker quits cleanly, every time: We did a top-to-bottom pass on shutdown so quitting FileMaker no longer hangs waiting on the plug-in. Background workers, an in-flight chat reply, storage, and even a stuck server restart now all wind down within a bounded time instead of wedging the quit, and the plug-in checks for an unsaved Script Workspace tab before any FileMaker action rather than firing a gesture into a modal.
• Even more script steps round-trip faithfully: We kept comparing steps against a live copy of FileMaker and fixed more stragglers, including Set Field into a specific repetition, Add Account keeping its privilege set, and steps whose text held escaped quotes or a stray option keyword that used to clobber the real value.
• Script edits are harder to get wrong: An edit now verifies the step count after saving and calls out a trailing delete that didn't land instead of reporting a clean success, refuses a write when the script name it was handed doesn't match the tab it's on, and stopped raising a false "some steps changed" warning when nothing actually did.
• A sharper sense of what an agent is looking at: The chat window now surfaces the pieces of your current FileMaker environment it can pull into a turn as identifiers you can switch on or off, each with a small glyph for what it is, and an agent can pull in a relevant piece you left off when it needs it. Nudges about a stale solution now key off the file itself, so they stop pointing at the wrong file when several are open.
• Cleaner, more honest chat: Canceling a reply now keeps the reasoning and text that already streamed in instead of wiping it, a response that quietly stops streaming times out into a clear notice you can act on, resuming an interrupted analysis remembers the question you originally asked, and malformed tool calls from smaller local models get tidied away rather than dumped into the chat.
• Opus 5 leads the Claude menu: Opus 5 is now the front-and-center choice for the Claude provider, with earlier models tucked behind More.
• Reliability and polish: A FileMaker file that isn't allowed to run the plug-in is now reported as blocked rather than a vague failure, a background database operation no longer parks the bridge and wedges an edit, and a file that fails to upload now tells you why with a size check up front.

v1.7.2 · August 14, 2026

• Scripts are much harder to damage now: We put a lot of work into the case where an agent rewrites a whole script. If a rewrite would have collapsed or emptied a script, the plug-in now catches it, refuses it, and rolls back to the version you started with instead of committing the damage. Every edit also proves which window it's typing into at the moment it types, and any edit that can't name the file it belongs to is refused rather than fired blind, so a drifted Script Workspace tab can no longer catch a write meant for another script. Two agents editing the same script no longer step on each other, a rewrite that quietly landed as a no-op is now called out as the failure it is, and a whole-script change the model forgot to actually paste gets applied properly rather than silently doing nothing.
• Multi-step script edits are faster and more reliable: When an agent makes several changes to one script, the chat panel now bundles them into a single request that navigates straight to the script, verifies it arrived, and applies every change together, instead of a chain of separate round-trips that could drift apart midway. An agent can also hand over the script's id, not just its name, so the edit lands on exactly the right one.
• Even more script steps round-trip faithfully: We kept comparing steps against a live copy of FileMaker and fixed the stragglers, including Show Custom Dialog written in FileMaker's own compact form, Insert Text, several Find-related steps whose options came back in the wrong order, blank comment lines, and layout references that were storing the label where the name goes. Steps whose settings are a set of checkboxes now match FileMaker exactly.
• Cleaner answers, especially from local models: While a reply streams in, internal plumbing and tool-call scaffolding no longer leak into the text you read. Malformed tool calls from smaller local models are now repaired on the fly so those models actually get their work done instead of stalling, and code you paste into your own chat messages stays readable.
• A heads-up when another AI tool overlaps AgenticFM: If you've got a separate FileMaker-oriented MCP server connected that does some of the same jobs the plug-in already does, an agent now notices the overlap and can point it out, so two tools aren't quietly fighting over the same task.
• Preview tabs say they're previews: A script tab that an agent opened just to show you something now carries a clear Previewing indicator, so you can tell at a glance it's a look, not a live edit.
• Navigation double-checks it actually arrived: Jumps to a table, field, or layout now confirm they landed on the right thing before reporting success, and names wrapped in FileMaker 26's curly quotes resolve correctly instead of quietly missing.
• Reliability and polish: Fixed a crash that could hit while the chat panel was being re-created after an update, a hung request now surfaces as a clear error notice you can act on, and the installer gained a manual-install option alongside a refreshed install script.

v1.7.1 · August 4, 2026

• Script steps are a lot more reliable: We reworked the step catalog behind the scenes. Steps an agent writes now come back matching what FileMaker itself would build, and a long list of steps that used to quietly lose a setting round-trip cleanly.
• Swap the icons on a button or a button bar: Ask an agent to restyle a copied button or button bar and it swaps in icons from whichever library you name. Open Settings and click Open Skills Folder to find the icon-swap skill, which lists the collections that are ready to go, including Lucide, Heroicons, Tabler, Phosphor, Bootstrap, Boxicons, and Material Design. Any other icon collection on the web works too, as long as an agent can reach it.
• AgenticFM now works on non-English FileMaker: If your FileMaker interface is in Spanish, German, Japanese, or any other language, the plug-in used to read an open Script Workspace as closed and every script edit dead-ended. It now learns the localized window names from FileMaker's own menus at runtime, and the same fix covers the Data Viewer, Script Debugger, and Inspector. No language list to maintain, so a locale we've never seen works too.
• Your untouched steps survive an edit: When an agent rewrites a script, it normally replaces the whole thing, which meant a step you never asked it to touch could quietly come back missing an option. AgenticFM now snapshots the original from FileMaker and splices your unchanged steps back in exactly as they were, only recompiling the ones that actually changed. It also tells you how many it preserved.
• Script insurance now covers deletions: The safety snapshot only fired when steps were being inserted, so a rewrite that deleted every step first ran completely uninsured. Deleting is the more destructive half, and it now gets at least the same protection, with every skipped capture saying out loud why it skipped instead of staying silent.
• Edits land in the script you meant: If Script Workspace drifted to a different tab between an agent navigating somewhere and writing there, the edit went into the wrong script and still reported success. An agent can now state which script it believes it's editing, and the plug-in re-checks the live tab and refuses the write on a mismatch rather than damaging the wrong one.
• A closed Script Workspace fixes itself: Asking an agent to edit a script with Script Workspace closed used to be a silent dead end, and it would burn a dozen turns guessing at a recovery. It now opens the workspace for you and gets on with the edit.
• Big solutions feel a lot better: Long script inserts got dramatically faster (a 200-plus-step append used to slow down the longer the script grew), and background scanning now backs off while you're actively typing, so the beach-balls people saw on multi-gigabyte hosted solutions are gone.
• Long agent turns stay honest: A request that legitimately ran six or seven minutes would freeze its status line partway through and then falsely report "Server stopped responding", even though the work finished fine. The live timer and phase now stay visible for the whole turn, and the plug-in only calls it a failure when the server actually stops answering.
• Better questions, clearer results: When an agent asks you something, the card renders properly whichever shape it uses, and every question now ends with a "Provide more information" row so you can answer something that isn't on the list. If a request hits the query limit, you get a plain explanation and a nudge to ask something narrower instead of a scary "this version doesn't support that" message.
• Searching your solution now looks inside your scripts: Text search was quietly returning zero matches for anything living in a script body, even when the plug-in had that script indexed. It searches script bodies for real now, and says so when it genuinely can't.
• Analyze tells you what's actually wrong: Per-file Analyze used to fail with an impossible "bring it to the front and try again" for files that can never be fronted. It now names the real reason: the bridge isn't installed (with a button to fix it), a script tab needs saving first, or FileMaker is briefly busy. The Solution Intelligence cache badge also stopped raising alarms about a cache that's perfectly complete.
• Saved clipboard snippets survive a restart: A snippet you'd saved showed up in your history after relaunching FileMaker but couldn't actually be used. They now rehydrate on demand, so promoting, patching, or inspecting any snippet from your history works no matter how long ago you saved it.
• Reliability and polish: Fixed a crash while learning FileMaker's menu names, and a startup crash triggered by a third-party audio plug-in. Reading a script from an outside agent now reports line numbers that match what Script Workspace shows. The "update available" nag stopped appearing when you're already on the latest version. Settings gained a Show button for conventions.json, an expired trial can now request a reset right from the chat, and we retired the last of the fancy dashes from the interface text.

v1.6.0 · July 23, 2026

• Local model support with LM Studio, Ollama & oMLX: AgenticFM now talks directly to models running on your own machine, no cloud account or CLI needed. Point it at LM Studio or Ollama and it lists their live models on the spot, prompts you to start the server when it's not running, and streams answers straight into the chat. A generalized OpenAI-compatible connection means new local backends are a quick add, not a rebuild.
• Solution Intelligence updates: When you open a solution, a designation dialog lets you confirm up front which of the open files belong together, so analysis and navigation always target the right one. You can add a file to a solution by hand from a "+" picker, move a file from one solution to another, and mark a file as a reference so it's known but never treated as part of the solution. Under the hood a new canonical registry keeps every open file pinned to the solution that actually owns it, which clears up a whole class of "wrong solution" mix-ups when several are open at once.
• Pinned chat control: A new pin affordance in the status bar locks the conversation to a specific file. If you click away to another file mid-chat, AgenticFM flags it and brings the panel back so an action can't quietly land in the wrong place.
• Snappier reaction when you switch files: The plug-in now notices you've moved to a different FileMaker file the instant it happens, so the chat panel re-orients to the file you're actually looking at without the old lag.
• Open Quickly keeps working under custom menus: If the layout you're on uses a custom menu set that hides File > Open Quickly, AgenticFM now restores the standard menus just long enough to navigate, so jumps to tables, fields, layouts, and scripts keep working instead of dead-ending.
• Chat UI polish: Assistant messages now run full width with a soft scroll fade at the seam, in-dropdown notices share one consistent look, and the solution dropdown got a tidier two-line header and cleaner row icons. Question cards from an agent render their options properly, dollar-sign globals like $$VAR show literally instead of collapsing, and notice buttons and spinners are styled consistently throughout.
• Clearer failures instead of bare codes: When a command-line provider fails, you now see its plain-text reason rather than a naked exit number, with auth problems called out per provider. If your pinned model can't run, AgenticFM substitutes a working one and tells you rather than failing the whole turn.
• Reliability and crash fixes: Fixed a preferences read that could abort FileMaker on a bad value, hardened the plug-in's background worker threads against uncaught crashes (including one triggered by a third-party audio plug-in at startup), and corrected file identity so open-file and environment reads report the real file even in data-separation setups.
• Read and edit scripts across multiple solutions: With several analyzed solutions open, script reads and edits now resolve to the file that owns the script, and reaching a hidden helper window (the kind FileMaker opens behind the scenes for related files) now works through the Window menu.

v1.5.1 · July 15, 2026

• Ask an agent to build you a theme: AgenticFM can now generate a full FileMaker theme for you from a description and install it straight into your Themes folder, painting real fills, borders, corner radius, and text colors rather than a flat placeholder. Describe the look you want and it maps that onto a proper theme spec, ready to pick in Layout mode.
• Layout snippets now paste with their real appearance: When an agent copies field and object placements onto a layout, they now come across with their actual fills, borders, rounded corners, and text color instead of a see-through theme fallback, so what you paste looks like what you meant.
• Set reasoning effort per provider from the model picker: Each provider now has an "Options…" row that opens a small dialog where you pick how hard it thinks, from the levels that provider actually supports. Your choice sticks, shows up in the connected notice, and providers that ignore effort (like Gemini and Cursor) show the control greyed out so you're never guessing. We also fixed a tall provider list running off the top of the panel.
• Latest GPT-5.6 models lead the Codex menu: GPT-5.6 Luna, Sol, and Terra are now the front-and-center choices for the Codex and OpenAI provider, with the older 5.5 and 5.4 models tucked under More, and Sol is the new default to match Codex's own.
• Each script you jump to opens in its own tab: Asking an agent to open a script and everything it calls used to collapse into one reused preview tab that kept getting replaced, so you ended up with only the last one. Every navigated script now lands in its own persistent tab, and Open Quickly starts scoped to the current file for faster, more accurate jumps, reaching for All Files only when the target really lives in another open solution.
• FileMaker comes to the front before an agent acts: If you clicked away to another app mid-request, an action could land in the wrong place and only half finish. AgenticFM now brings FileMaker forward before every edit, run, or navigation, so your changes always go where you expect.
• Run scripts on FileMaker Server: The bridge gained a Perform Script on Server command, so an agent can kick off a server-side script and wait for its result, right alongside the existing run-a-local-script support.
• Better handling for solutions with a dot in their file name: A file named something like "AG-FM Inventory v1.4.1" was invisible to the plug-in's name lookups because FileMaker treats the trailing ".1" as an extension. Calculation-based lookups against the frontmost file now resolve it correctly. Running scripts in such a file still isn't possible, though: FileMaker itself refuses to start a script by a name that contains a dot, and there's no workaround for it, so you now get a clear error saying so instead of a silent miss. If you need an agent to run scripts in a file, avoid dots in its name.
• Claude no longer gives up mid-task: On multi-step requests Claude could run out of turns before it even reached your solution and quit with an error. We raised its turn ceiling so routine "go find this script and show it" requests finish, and when it does hit the limit you now get an amber notice with a one-click Continue button that picks right back up where it stopped.
• A heads-up when two copies of FileMaker share the plug-in: Running two FileMaker processes off the same installed plug-in used to quietly cause port collisions and folder contention with no warning. You now get a clear, dismissible notice listing every FileMaker that's sharing it, so a silently unreachable panel isn't a mystery anymore.
• Attach files to a bug report: The Send Support Bundle dialog now lets you attach screenshots or sample files by button or drag-and-drop, so you can show exactly what you're seeing. Attachments are size-capped with a friendly nudge to share a download link if they're too big, and they're signed right into the bundle.
• A new Script Workspace chat button toggle: Settings > Appearance now has a switch to show or hide the AgenticFM button in the Script Workspace toolbar, and it applies live without relaunching FileMaker.
• Script edits stop politely when a dialog is in the way: If FileMaker has a modal prompt up (like "you have scripts with unsaved changes"), an agent's script edits used to fire keystrokes into the void and fail with a confusing message. They now detect the dialog, stop, and tell you to dismiss it first, and edits driven from the chat panel now reliably land in Script Workspace even when another window had focus.
• Script bodies load reliably across multiple solutions: With more than one analyzed solution open, or right after a restart, asking an agent to read or edit a script could come back empty even though the script was fully saved. Scripts now hydrate from the solution that actually owns them, so read and edit work everywhere, and an empty body now says so loudly instead of pretending it worked.
• No more invented "lives in" locations: When a script couldn't be traced to any analyzed file, the plug-in used to name an arbitrary file as its home, which an agent would then repeat as fact. It now says honestly that the location is undetermined instead of guessing, and only names a file when a real reference points there.
• Clearer bridge notices: The post-upgrade "other files still need the bridge update" prompt was firing on every solution by mistake and sticking around after you switched files; it now only appears when it's actually true, is dismissible, and clears itself when you move on. And a bridge that's installed but out of date is now correctly flagged as needing a re-copy instead of showing up as "not installed."
• Chat history, previews, and polish: Past conversations restore their full transcript correctly and stray internal markup no longer leaks into the history list or the copy-as-text output. Inline code inside your own chat bubbles is readable again, the chat panel refreshes faster with no history flash when you switch files, rendered previews gained a Back-to-list button, and every error notice now offers Send Support Bundle from one consistent place. Also fixed a crash that could hit while pinning script tabs and a bridge version misread for comma-decimal locales.

v1.5.0 · July 6, 2026

• Skills are here! Agents can use skills: AgenticFM now has a Skills folder. Drop a skill into it, and the agent discovers it on its own and follows it when it's the right fit, reading the details only when it actually needs them. Turn the whole thing on with a new Skills setting (off by default), and there's a one-click button to open the folder so you can add your own.
• Give agents access to folders: A new folder button in the status bar lets you hand the agent specific folders on your disk to work with, so it can read your project files, assets, or exports right alongside your solution. Grants are scoped per solution and everything is read-only until you say otherwise, with a confirm step before you ever enable writing. Pick a folder with the native chooser, manage or remove grants anytime from the Manage sheet, and if your chosen provider can't fully honor read-only you get a plain warning saying so.
• Automate other apps with AppleScript: New Host Automation setting (Never, Ask, Always) decides whether the agent may run AppleScript on your Mac to drive other applications. It always shows you the script and its result first, and Never simply refuses so nothing runs behind your back.
• SQL INSERT, UPDATE, and DELETE, with real guardrails: Agents can insert, update, and delete records for you through safe parameterized queries, so your captured, web, or typed content can never sneak into the query itself. A new Data Modification control (Never, Ask, Always) decides when it's allowed, and any update or delete without a WHERE clause always stops to confirm first even on Always, because there is no undo for a data write.
• Enhanced clipboard operations: Copy a whole set of field placements onto a layout snippet, pulling the fields straight from a table, from your analyzed solution, or from another snippet you've copied. Match them up by order or by name, and let it add or trim placements when the counts don't line up. You can also ask it to list the source's fields first, pick just the ones you want, and it rebinds only those.
• See and control what the agent can reach: A new Agent Self-Configuration section in Settings lets the agent read and adjust its own configuration, and shows you a plain-English summary of exactly what it can read, what it can write, and what it can never touch, so nothing about its access is a mystery.
• Cleaner Settings pane and clipboard menu: We tidied up the Settings pane for better readability and made a handful of fixes to the clipboard menu.
• Latest Claude models in the picker: Fable 5, Opus 4.8, and Sonnet 5 now appear in the Claude provider menu, with older models tucked behind the More flyout, and Sonnet 5 is the new default.

v1.4.1 · July 1, 2026

• No more beach-balls on big solutions: Working in a large hosted solution with hundreds of tables and hundreds of thousands of fields used to freeze FileMaker for several seconds at a time while the plug-in inventoried every field, both when building context for a chat and on its own background refreshes. That heavy field scan is now skipped automatically on large solutions, so typing, clicking, and window switches stay smooth even mid-conversation.
• Faster, more reliable jumps to a specific script: Navigating to a script by name in a particular file now brings the right window and Script Workspace forward and lands on the exact script directly, instead of routing through Open Quickly.
• Heads-up when a custom menu set hides Open Quickly: If the layout you're on uses a custom menu set that leaves out File > Open Quickly, the plug-in now tells you up front that general navigation to tables, fields, layouts, and scripts is limited, rather than only failing when you try it. Scripts can still be reached directly by naming their file.

v1.4.0 · June 28, 2026

• Chats are now saved and you can pick up right where you left off: AgenticFM now keeps a history of your past conversations, scoped to the solution you're in. Open the history panel from the toolbar (rewind clock icon) or press Cmd-J to browse and filter your prior chats, then click any one to replay its full transcript read-only or hit "Continue session" to resume it live. New Chat History settings control how long history is kept, by age and by number of chats, with automatic cleanup so it never piles up.
• New Script Insurance, scripts get a safety net every time an agent writes one: Before an agent inserts or patches a script, AgenticFM quietly snapshots the original straight from FileMaker, then compares it to the result afterward. If any step came back changed in a way you didn't ask for, that step gets flagged so the agent can check it with you instead of silently letting an option get dropped. It runs automatically on both editing paths, with nothing for you to turn on.
• Every script step now round-trips correctly: We went through FileMaker's entire script-step library and verified each one against a live copy of FileMaker, fixing the steps that were losing settings on the way in or out. Show Custom Dialog now keeps all your buttons and input fields, Send Mail keeps its attachments and Cc/Bcc, and dozens of others (Open URL, Set Web Viewer, Truncate Table, Go to Related Record, Open and Close File, Convert and Replace File, Perform Script on Server with Callback, Perform Semantic Find, and more) now preserve every option. Practically anything the agent writes for you comes back faithful to what you'd build by hand.
• FileMaker 26 support for new script steps: Added support for FileMaker 26's newest steps, including Configure Persistent Data and the new AI insert steps, so the agent can read and write them just like the rest.
• Agents can ask a question right in the chat: When an agent needs to check something with you before continuing, its question now shows up as a proper question card you can answer inline, instead of plain text you had to puzzle over.
• Cleaner streaming answers: While a response was still streaming in, longer replies could briefly break into side-by-side columns. Streaming text now lays out in a single clean column the whole way through.

v1.3.3 · June 23, 2026

• An agent can now take screenshots: Ask it to grab a picture of a FileMaker window, the whole screen, or any other app's window and it captures it for you. Window captures are occlusion-proof, so they grab the real window even when something is sitting on top of it, including FileMaker's own dialogs. You can also kick off an interactive crosshair selection and drag out exactly the region you want. Each capture comes back as a card with a Preview button that opens the image in Preview.app. Future use will provide the image to the agent for vision processing.
• You decide when the agent may capture: A new Content Capture control in Agent Behavior lets you set capturing to Never, Ask, or Always, just like the other safety policies. Every Agent Behavior setting now also carries a short help line under it explaining what it does, so you are not guessing what each one controls.
• Save your favorite clipboard snippets and keep them across sessions: The clipboard bar now has a Saved section that sticks around after you quit FileMaker and rehydrates the next time you open the chat. Name a snippet, rename it anytime with the pencil, and star the ones you reach for most so they float to the top. Delete one or clear them all with a clear three-way prompt so nothing disappears by surprise.
• Script creation tells the truth: When an outside agent (Codex, Claude desktop, your own scripts) asks the plug-in to create a script, a create that did not actually land is now reported as a failure instead of a misleading success. New scripts also open as the frontmost Script Workspace tab and get named reliably, instead of occasionally committing as a generic "New Script".
• Navigate to exactly the script you mean: Jumping to a script by name used to fall back to a prefix match that could land on the wrong one when two scripts share a prefix or name. Navigation now targets the exact script every time, and expanding all script folders to find it is much faster.
• External agents learn about plug-in updates: If you drive the plug-in from an outside agent and never open the chat panel, you used to miss the "update available" notice entirely. The agent is now told when a newer plug-in version is pending so it can let you know and walk you to the in-panel installer.
• More accurate sense of which script is active: The plug-in now reads the live active Script Workspace tab rather than a slightly stale cached value, so actions aimed at "the script you're editing" land on the right one right after you switch tabs.
• Fixed steps that take a field or a variable: Steps like Get File Exists that accept either a field or a variable target no longer mangle a variable target into a broken field reference (showing "Table Missing") or a literal path. Variables now round-trip correctly in both directions.
• Crash fix: Fixed a crash that could occur during idle background work.

v1.3.2 · June 18, 2026

• External agents get the full Solution Intelligence treatment: When you drive AgenticFM from an outside agent (Codex, Claude desktop, or your own curl scripts), it now receives the same Solution Intelligence and Preview guidance the built-in chat already uses, including a first-contact orientation that points it at analyzing your solution before it starts editing. An external agent can also open the chat panel on its own, so you can watch a solution index in real time.
• CLI providers installed through nvm or pnpm now get found: If your Pi, Codex, or Gemini CLI lives in a Node version-manager folder, AgenticFM now reads your login shell's full PATH when probing and launching it, so a provider that used to come up empty or fail at runtime now resolves correctly.
• Long script names stay inside the chat panel: Solutions with very long script names (think 40 to 90 characters) could push the chat panel wider than its window and leave you scrolling sideways. The status bar, clipboard preview, and attachment chips now trim long names with an ellipsis and show the full name on hover, so the panel keeps its width no matter how your scripts are named.

v1.3.0 · June 14, 2026

• Pi Coding Agent joins your provider lineup: A new provider alongside Claude, Codex, Cursor, and Gemini. Pick it from the provider menu and tap into its super wide range of models (see the full list at https://pi.dev/models).
• Refresh providers without restarting FileMaker: Installed a new agent CLI after launch? A new Refresh row in the provider menu (with a spinning icon while it works) re-scans for newly installed tools and picks up their models on the spot, no FileMaker relaunch needed.
• Safer layout edits with built-in validation: Layout changes are checked against a catalog of rules before they apply, so a malformed edit gets flagged with a clear reason instead of quietly producing a broken object.
• Provider errors you can actually read: When an underlying CLI returns a raw error (often a wall of JSON), AgenticFM now lifts the human message out of it instead of showing the envelope. Unexpected failures offer a one-click Send Support Bundle, and error notices render full width to match the rest of the chat.
• Smoother first run for external agents: An outside agent (like Codex or Claude desktop) can now start a trial and bootstrap the plug-in on first run without anyone opening the chat window. The remote-access banner gained a "Copy agent instructions" button that hands over everything needed to connect in one paste, with Copy URL, Copy Token, and Regenerate tucked into a tidy menu.
• Failed script inserts can't masquerade as success: When a step insert fails, the failure is passed back to the agent so a follow-up "save" can't claim everything worked.
• More provider and installer fixes: Codex no longer drops your chosen model when resuming a conversation, Cursor's prompt handling regression is fixed, and the installer file now has a Launch button to open FileMaker right after installing.
• Ready for FileMaker 26: Solution analysis now reads exports correctly on FileMaker 26 (UTF-8) as well as 25 and earlier (UTF-16).
• Chat UI polish: The script preview pane is now collapsed by default and renamed "Validated Script", a new theme-blended banner helps you link an external agent running on the same computer.

v1.2.5 · June 13, 2026

• Clickable line citations: When the agent points at "Lines 17 to 22" of a script, those numbers are now a link. Click it and the script opens in Script Workspace with exactly those rows selected and scrolled into view, so you jump straight from the analysis to the code. The agent can also label the link with a step name ("the Show Custom Dialog step") instead of bare numbers.
• Script Workspace fixes: More fixes to the Script Workspace to make it more reliable and user-friendly.

v1.2.4 · June 10, 2026

• Beach-balls are for the beach, not for big hosted solutions: Idle background work on multi-gigabyte hosted files used to stall FileMaker for several seconds every 30 seconds, freezing typing and window switches. Background cache refreshes now stay out of the way when nothing is using them.
• Solution Intelligence freshness is now visible: When the cached analysis of a solution is older than your configured threshold, a clear notice appears in the chat with an Analyze button, and the same staleness signal is exposed on the agent's side so it can offer to refresh on its own. New Settings controls let you choose how old is too old and whether the plug-in warms Solution Intelligence on startup.
• Missing Solution Intelligence banner now fires when it should: When a chat session asked the agent about a solution that hadn't been analyzed yet, the suggestion to run Solution Intelligence sometimes stayed hidden because another solution was already loaded. The banner now checks the specific solution you're asking about, and a polished new spinner notice surfaces while the cache is loading instead of stretching it out behind a tiny refresh glyph.
• External agents can target a specific file again: Driving the plug-in from curl, Postman, or Codex against a specific file works again across all script operations (navigate, insert, delete, save, create). Open Quickly now also pre-opens Script Workspace correctly when it's closed and raises the right window when more than one Script Workspace is open.
• Clipboard writes from the chat no longer fail validation: Whole-script clipboard payloads were being checked against a rule meant only for step-shaped snippets and rejected with a confusing error. The validator is now class-aware, and the chat's Copy Script paths tag their payloads correctly so they go straight through.

v1.2.2 · May 27, 2026

• Snappier chat panel after a send: The chat now polls for completed responses on a shorter interval. A 5-minute watchdog now surfaces a clear 'Server stopped responding' notice instead of polling forever when something hangs.
• Codex CLI multi-turn conversations fix: Earlier builds dropped every prior assistant turn on the Codex provider, so each message looked like a fresh thread. Codex sessions now correctly resume via the official method, preserving the conversation across turns. Verified against Codex 0.130 through 0.134.
• More reliable surgical script edits: Multi-step patch batches no longer leave a script half-applied when a cached lookup goes stale between directives. The agent reads live Script Workspace state between operations and can also locate an open script through Script Workspace when Solution Intelligence has not yet indexed the file.
• 'Visualize this' requests render a real graph: Asking the agent to show, preview, diagram, visualize, or graph a dependency now produces an actual Mermaid graph instead of a markdown table. A 'Render as graph' notice appears as a fail-safe if the agent forgets, and the same intent is now recognized in German, Japanese, French, Spanish, and Italian as well as English.
• Script names with curly quotes resolve correctly: When the agent emitted a script reference wrapped in smart quotes (“Name”) instead of straight quotes, lookup quietly missed and surfaced a confusing 'couldn't find this script' card. Straight and curly quotes are now both stripped before resolution.
• Clipboard stability: Reading and writing the clipboard now happens on the main thread the way macOS requires, fixing a crash that could hit when the chat panel, bridge, and a running script all touched the clipboard at once.

v1.2.1 · May 26, 2026

• Surgical script edits with new patching system: When an agent needs to tweak a script, it can now emit a precise patch (replace this step, insert these after step N, delete from M to K, append at the end) instead of regenerating the whole thing.
• Faster multi-turn chats via prompt caching: System prompt assembly has been updated and now follow-up turns return faster and cost fewer tokens.
• Attachment chips above prompt input: Clipboard contents and the script you're editing in Script Workspace now appear as chips above the input bar, each labeled with its size. Click a chip to exclude that attachment from the next turn; the following send re-enables everything by default.
• Faster, more targeted Solution Intelligence queries: An agent can now ask for compact entity summary cards, a step range from a single script, filtered reference and dependency lists (for example, only scripts that run with full access or use plug-in steps), and structurally aware text-search hits that say exactly which slot of which container matched.
• Adjustable correction attempts: Settings → Validation now has a control for how many automatic correction passes the agent runs after each response (0, 1, or 2). Default unchanged.
• Pre-flight prompt-budget check on huge solutions: Very large schemas (hundreds of tables) no longer trigger a cryptic "argument list too long" failure. Over-budget prompts are now rejected locally in under 100 ms with a clear message showing the estimated token count and the model's context window.
• Security and reliability hardening: Allow and deny security endpoints restricts an agent from using potentially dangerous tools like shell-capable plug-ins and binaries. Click the Configure button for the AgenticFM plug-in within FileMaker's settings to access security-allow.json and security-deny.json.

v1.1.1 · May 22, 2026

• Window-aware safety net for live actions: When the agent runs a script, edits one in Script Workspace, or evaluates a calc, AgenticFM now checks that you're still on the file the conversation started in. A new "Target File Switching" preference (Ask, Always, Never) decides what happens when you've switched files mid-chat. Pick Ask and you get a one-click "Go back" or "Rebind" notice. Pick Always and the agent quietly re-fronts the bound file before running. No more actions firing into the wrong file.
• Cursor Agent CLI joins the provider lineup: Compose with Cursor's headless CLI alongside Claude, Gemini, and Codex. Composer-2.5 is the seeded default, and the full Composer catalog plus access-gated third-party models (Sonnet, Opus, Haiku, GPT-5.x, Gemini 3.1 Pro, Grok 4.3, Kimi K2.5) ship out of the box.
• Built-in help and slash-command picker: Type "help" (or the localized equivalent in DE, ES, FR, IT, JA) and press Enter to open the full help page right inside the chat. Type "/" alone to bring up a floating picker of available commands with arrow-key navigation. An always-visible hint below the input bar surfaces the keyword so new users can find it without docs spelunking.
• Cleaner model picker: Each provider's model menu now shows current-generation models inline and tucks legacy entries behind a "More…" flyout, so dropdowns stay short.
• Customizable script header: Drop your preferred Purpose / Parameters / History banner into conventions.json and the agent uses it as the opening of every generated script. Multi-line headers, placeholder tokens like {purpose}, {author}, {version}, and {date}, and your existing default all work without any new settings pane.
• Post-insert script review: After the agent inserts or creates a script, AgenticFM walks the rendered result and flags anything that landed as <unknown> or <Table Missing>. You get a flat list of "Step N: …" notices identifying the exact missing layout, script, field, or table occurrence. Layout references now also emit cleanly (no stale id=0).
• Close Window safety rule: Generated scripts will no longer quietly close your work window. A new prompt rule plus a verifier flag any standalone Close Window step that isn't paired with a New Window earlier in the same script. Asking the agent to "close quietly" now maps to Commit Records + Exit Script instead.
• Anti-hallucination guard for fake tool names: When the LLM invents a tool that doesn't exist (or fabricates a "permission denied" error to cover for it), AgenticFM intercepts the response and surfaces a clear notice. Driven by a catalog, so future patterns are a JSON update rather than a rebuild.
• Insert again at current caret: After a successful script insert, the success card now keeps an "Insert again at current caret" button. Re-paste the same block at your cursor without re-prompting the LLM.
• Robust file and window targeting: AgenticFM now sees every open FM file via DatabaseNames, including kiosk-style files with no visible window and two copies of the same file open at once. Window-targeting steps (Select Window, Close Window, New Window) now use FileMaker's canonical WindowNames internally, so they reliably hit the right window instead of silently no-opping on titles that carry a host suffix. Empty scripts in Script Workspace no longer surface FM's placeholder row as a phantom step.
• Script Workspace line numbers now match what you see: Inserts land at the line you targeted. Line numbers in the prompt are 1-based throughout, matching the number Script Workspace shows next to each step, so afterIndex: 4 puts the new step right after step 4.

v1.0.0 · May 15, 2026

• One-click in-app updates: When a new version drops, the update banner now installs it for you. The new build downloads, gets verified, and the bundled updater file opens automatically. Just relaunch FileMaker when prompted.
• Warm starts feel instant: Plug-in launch time dropped from around 14 seconds to about 35 milliseconds. More aggressive startup caching, so FileMaker stays responsive the moment the plug-in loads.
• Polish pass on chat animations and layout: New send-button arrow loop on every send, a subtle wiggle when the context lock toggles, a breathe animation on New Chat, plus a stack of layout fixes (clipboard menu overflow, code-block copy buttons, gallery row wrapping).
• Smarter solution attribution for runtime file paths: When a frontmost file resolves at runtime from a $$VAR external data source, the chat now attributes it to the right solution and offers a one-click Analyze. The dropdown shows the runtime mapping (for example, "Data File → Solution File") instead of leaving you guessing.
• Cleaner solution files dropdown: External data source cascades collapse into a single row per logical reference rather than one row per fallback path. No more phantom "missing" entries when an EDS lists $$VAR fallbacks alongside filemac alternates.
• Larger solutions in Solution Intelligence: Maximum SaXML size is now configurable in the Solution Intelligence Cache pane (default 800 MB, up to 4 GB). Files rejected for size tell you the current cap and the actual file size so you know what to bump.
• Appearance mode moved into the theme picker: System, Light, and Dark sit in a compact icon row at the top of the theme dropdown. Picking a theme that doesn't support the current mode falls back to a supported one in place rather than stranding the UI.
• Trial countdown is consistent everywhere: The trial banner and Settings now read the same number of days remaining, and a fresh 14-day trial actually says 14 until a full calendar day has elapsed.
• Crash hardening: Fixed a crash when polling the Data Viewer while a script was paused on a Set Variable step. Fixed a separate trial-activation crash when a slow network request returned after the wait had timed out.
• Security and release hardening: Tightened the localhost Origin allowlist, added path-traversal guards on theme IDs, SHA-256 verification of downloaded updates, stricter notarization checks on release builds, and license-key redaction in support bundles.

v0.10.0 · May 6, 2026

• Solution Intelligence now persisted to disk: Reopening a previously analyzed file restores its full analysis in milliseconds without re-parsing the source. Reanalyze easily for up-to-date data.
• Auto-load on file switch: Opening a known file shows a one-shot "Prior Solution Intelligence loaded" notice with a Reanalyze button; can be disabled in Settings.
• New Solution Intelligence Cache pane in Settings: Per-solution disk usage and freshness details, individual delete or clear-all, and a configurable storage budget with automatic eviction of the least-recently-used solutions.
• Brain-glyph count badge: Showing the number of loaded files color-coded for freshness state.
• Solution Intelligence dropdown: Now lists EDS-referenced files that have never been analyzed in a collapsed group, each with a one-click Analyze button, plus a ✕ icon to remove a file from the solution (persisted across sessions).
• Seven new agent analysis queries: Plug-in usage, folder analysis, spelling drift, duplicate detection, file-access permissions, local-variable inspection, and rule evaluation.
• Clipboard intelligence: The agent can read your FileMaker clipboard, digest its contents into context, and apply targeted patches back. Clipboard history shows size-aware entries with wall-clock timestamps; click anywhere on the clipboard bar to open the menu.
• Reasoning-effort knob per provider: Set Claude or Codex reasoning effort once in your provider config; it surfaces in the welcome banner each session.
• macOS keybindings permission self-heals: Once you grant Accessibility or Input Monitoring access, copy/paste shortcuts work without restarting FileMaker, and a one-click button deep-links to the right Privacy pane.

v0.9.2 · April 22, 2026

• AI-powered script generation and modification from natural language
• Automatic HR-to-XML conversion with validation
• Multi-provider support (Claude, Gemini, GPT)
• Native floating chat window with theme support
• Clipboard-native fmxmlsnippet output
• Solution-aware context engine