FLASH CLI v0.5.6

What's Changed

• Add Settings to Flash Web: open it from the bottom of the sidebar or with Alt ,. It has three tabs: General, Usage, and Memory by @Natuworkguy
• See your usage at a glance: messages sent, tokens generated, tool calls, time working, active days, your current and longest streak, and average speed, with a year of activity as a heatmap and each model's share of your turns by @Natuworkguy
• Import memory from another AI: copy the prompt Flash gives you into ChatGPT, Claude, Gemini, or any other assistant, then paste its reply back. Each fact goes straight into memory as written, with no model in between, and you see what is new before anything is saved by @Natuworkguy
• Edit memory in the web UI: add a fact, change one in place, or forget it. Changes reach the model on the next message, without a restart by @Natuworkguy
• Pick Light, Dark, or System under Appearance. System follows your computer as it switches, and every change sweeps across the screen from the top-left corner by @Natuworkguy
• Start a chat in a project from the message box: the Project button on a new chat lists your projects, and the chat starts in whichever you pick, or you can make a new project from the same menu by @Natuworkguy
• Rename a chat from the sidebar: hover it and click the pencil, or double-click it. A chat in a project now shows the project's name beside its title by @Natuworkguy
• Add the Flash bolt as the tab icon and beside the name in the sidebar by @Natuworkguy
• Fix sub-agents in Flash Web: when a sub-agent finishes, the chat that started it picks up what it found and replies on its own, the way the terminal does. A sub-agent's news only goes to its own chat, and a chat is woken at most three times in a row by @Natuworkguy
• Show running sub-agents in the web UI: a "1 agent working" button beside the model picker lists each one with its task, how long it has run, and what it is doing by @Natuworkguy
• Fix a web turn that started a sub-agent ending in a TypeError instead of showing its token count and time by @Natuworkguy
• Draw plans in the web UI as a checklist that ticks off as the model works, instead of printing them in the terminal running the server. Waiting on a sub-agent no longer draws in that terminal either by @Natuworkguy

Full Changelog: v0.5.5...v0.5.6

FLASH CLI v0.5.5

What's Changed

• Add Flash Web: run flash --web to use Flash in your browser. It's the same agent as the terminal, with the same model, tools, permission prompts, skills, and extensions, and replies stream in as the model writes them. Tool calls fold into rows that expand, file changes show as colored diffs, and a permission prompt takes the keyboard so a single y or n answers it by @Natuworkguy
• Drive the web UI from the keyboard: Ctrl K searches every action, chat, and model, and ? lists every shortcut. Each button shows its shortcut when you hover over it by @Natuworkguy
• Open Flash on your phone: flash --web --lan listens on your network and prints a QR code to scan, and Open on phone in the sidebar shows the same code. Every request needs the random token in that link, and a reload keeps you signed in with a cookie no other site can use by @Natuworkguy
• Add /web to start the web UI from inside a terminal session, /web lan to open it to your network, and /web stop to shut it down by @Natuworkguy
• Add projects: a project is a folder plus instructions. Every chat in it runs its tools in that folder with the instructions in its system prompt. Web chats are now saved, so a project keeps its conversations across restarts by @Natuworkguy
• Switch Ollama hosts from the model menu: add another machine, see which hosts are answering, and pick one to list its models by @Natuworkguy
• Add chat search: Alt F searches every chat's titles and messages, and opening a result scrolls to the message that matched with your words highlighted by @Natuworkguy
• Add a send_pdf tool, and show sent files in the web UI: images and PDFs slide out in a panel on the right with the file name and size across the top, and a card in the chat opens them again. In the terminal, a PDF opens in your PDF viewer by @Natuworkguy
• Add a learning loop: Flash saves how to do tasks that come up again as skills in ~/.flash/skills, reads the matching one before it starts, and puts your saved memory in its system prompt. Every 10 tool calls or 10 messages, a background review saves what was worth keeping, and it only ever changes skills Flash wrote itself. /skills lists what it has learned by @Natuworkguy
• Add extensions: flash --extension-install github@username/my-ext or /extension install adds slash commands, tools the model can call, system prompt text, and backgrounds from a GitHub repo, after showing you what it adds and asking. See docs/EXTENSIONS.md to write one by @Natuworkguy
• Add precise file edits: the edit and multi_edit tools change a file by naming the exact text that changes instead of rewriting it, so a one-line fix in a big file costs one line of output. Several changes to one file land together or not at all by @Natuworkguy
• Add /undo to put back every file the last turn changed, including deleting the ones it created by @Natuworkguy
• Measure history against the model's real context window, and summarize what falls off instead of dropping it, so a long session keeps its thread. /context shows the usage and /compact summarizes on demand by @Natuworkguy
• Add terminal keys: Alt+Enter or a trailing \ starts a new line, Up and Ctrl+R bring back earlier messages saved across sessions, Shift+Tab toggles autonomous mode, and Ctrl+O shows cut-short tool output in full by @Natuworkguy
• Cut long tool output to its first five lines, or its last five when a command fails, so a test run no longer pushes the reply off the screen. The model always sees all of it by @Natuworkguy
• Frame the prompt with a status line showing the model, the host, how full the context is, running sub-agents, and a dot for whether Ollama is answering by @Natuworkguy
• Close each turn with how long it took, like "Pondered for 12s · done 14:02" by @Natuworkguy
• Add pixel-art backgrounds behind the prompt: /background lists the scenes and /background <name> switches. Drop your own .scene files in ~/.flash/backgrounds by @Natuworkguy
• Complete emoji as you type: :fire offers 🔥 by @Natuworkguy
• Keep the screen intact when you resize the terminal: the banner and the conversation redraw at the new size instead of breaking apart by @Natuworkguy
• Clear edits out of VS Code once you've answered them, instead of leaving the diff looking like it's still waiting. Set workbench.editor.closeOnFileDelete to true to have the tab close too by @Natuworkguy
• Fix edits on Windows: an edit spanning more than one line now matches files with Windows line endings, and the file keeps them by @Natuworkguy

Development

• Add a ruff configuration covering import order, pyupgrade (except the two annotation rules, so Union and Optional stay), blind excepts, and naive datetimes by @Natuworkguy

Full Changelog: v0.5.4...v0.5.5

FLASH CLI v0.5.4

What's Changed

• Add async sub-agents: the model hands independent pieces of a task to sub-agents with the agent tool, lets them run in the background, and ends its turn. When one finishes, Flash wakes the model with the answer so it reports back without you typing anything. It only wakes while your prompt is empty, never takes a half-typed message, and stops after three wakes in a row by @Natuworkguy
• Add agent_result for when the model can't go on without an answer: it waits for the sub-agent and draws its progress live, with the current step, the round, the time so far, and its latest tool calls by @Natuworkguy
• Add /agents to watch every sub-agent update in place, and /agents <id> to see one in full with its answer. Sub-agents keep running after a reply, and a dim line under the reply says how many are still going by @Natuworkguy
• Sub-agents get the file, search, and web tools, plus shell and write in autonomous mode only, since they have no terminal to ask from. Their tool output stays in their progress view instead of landing in your terminal, and they ask Ollama for the same context window as the main loop, so the model never reloads between them by @Natuworkguy
• The model now sees commands you run with !: the command, its exit code, and its output ride along with your next message, so "why did that fail?" works right after one by @Natuworkguy
• Add /hook install for VS Code: a small zsh or bash hook, active in VS Code's terminal only, records the commands you run there and their exit codes, and Flash attaches the ones since your last message. The model knows which command failed and where without you pasting anything. The hook sees exit codes but not output, so the model re-runs a safe build or test to read the error and asks you to paste it for anything that deploys. Commands that start with a space are skipped, secrets are redacted, and /hook remove takes it back out by @Natuworkguy
• Work with VS Code when Flash runs in its terminal: a file edit waiting for your yes opens as a side-by-side diff in the editor, and a new open_in_editor tool opens a file at the line the model is talking about by @Natuworkguy
• Add visible plans: for a multi-step task the model posts a checklist with the plan tool and ticks each box with check_step as that step lands. /plan reprints the current list by @Natuworkguy
• Add a send_image tool: the model hands you a finished picture, drawn inline in terminals that support images and opened in your image viewer everywhere else by @Natuworkguy
• Render LaTeX math in replies as Unicode, so \frac{\pi}{2} reads as π/2 instead of arriving as TeX source by @Natuworkguy
• Fix /update on Windows: the update now installs once Flash has exited, since a running install can't replace itself there by @Natuworkguy
• Keep voice mode working across updates: /update puts the voice packages back, and install.sh and install.ps1 force-reinstall them so they actually move to the new version by @Natuworkguy

Flash Onyx 2.5

• Double the context window to 131,072 tokens by @Natuworkguy
• Give up the name outside Flash: the assistant has no name of its own and goes by whatever the app running it calls it, which is Flash in the Flash CLI. Asked what it is, it says it's the Flash Onyx 2.5 model on a Gemma base by @Natuworkguy
• Cover far more ground in code: new guidance for JavaScript and TypeScript, React, Go and Rust, SQL, regular expressions, algorithms, HTTP APIs, web security, systems and networking, builds and CI, data work, charts, spreadsheets, and machine learning by @Natuworkguy
• Go beyond code: new guidance for health, money, business, translation, tutoring, creative writing, science, puzzles, dates and time zones, everyday asks, when to say no, and people in a bad place by @Natuworkguy
• Treat everything it reads as data: a file, page, issue, or search result that tries to give orders gets quoted back to you instead of obeyed by @Natuworkguy
• Handle tool failures deliberately: read the error, tell missing from not permitted from malformed from refused, and say which instead of quietly working around it by @Natuworkguy
• Better at code and at sounding human, with tighter voice-mode replies and the working shown before any computed number by @Natuworkguy
• Add music theory and MIDIUtil guidance, and rules for rendering images in Python by @Natuworkguy
• Make good use of whatever the app running it provides, each rule applying only where that feature exists: start sub-agents for independent work and end the turn instead of waiting on them, answer from a failed command's output when it's there, never re-run a deploy even after reading its script, and open a file at the line when asked to show where something is by @Natuworkguy

Tip

Use Natuworkguy/flash-onyx-2.5:31b-cloudbase to try the cloud build, or flash-onyx-2.5:12b on consumer hardware.

Model builds

• Write generated Modelfiles with \n line endings on every platform in models/build.py by @Natuworkguy

Full Changelog: v0.4.4...v0.5.4

FLASH CLI v0.4.4

What's Changed

• Add an interactive model picker: bare /model opens an arrow-key list of the models on this machine, typing filters it, and a name that matches nothing is offered as a download with a progress bar. /model <name> still switches straight to one by @Natuworkguy
• Add @ file mentions: type @ anywhere in a message to pick a file from a dropdown, use / to walk into a directory, and the model reads whatever you pointed at. Dot-entries stay hidden until you type the leading dot by @Natuworkguy
• Add a fetch tool that reads a web page as text with no browser and no vision model. It returns the final URL, title, description, and body, capped at 20,000 characters, and hands JSON, plain text, and CSV back untouched. A page drawn by JavaScript comes back with a note that tells the model to screenshot it instead by @Natuworkguy
• Teach the model when to read and when to look: fetch for what a page says, screenshot for how it looks. Search results are now fetched before being relied on, rather than answered from a snippet by @Natuworkguy
• Add SHOW_STATS: a dim line under every reply with the tokens the turn used, how long it took, the generation rate, and how full the context got. Tool rounds are summed into one line, and SHOW_STATS=0 hides it by @Natuworkguy
• Add NUM_CTX: set a token count, or max for the largest window the model's architecture supports. Flash says once when the setting did nothing and why, and once per model when no window is pinned and Ollama's default applies, with the ceiling and how to raise it by @Natuworkguy
• Fix voice mode speaking replies to typed messages. Only a turn you spoke gets a spoken reply; a typed turn is answered in writing and hands the prompt back by @Natuworkguy
• Name the file path in the write tool's result line by @Natuworkguy
• Add httpx to the requirements. install.sh and install.ps1 pull it in; a manual install needs pip install -r requirements.txt again by @Natuworkguy

Flash Onyx 2.4

• Put the steps before a computed answer: a number or verdict it still has to work out comes after the arithmetic and a one-line check, instead of being guessed first and justified after by @Natuworkguy
• Stop fabricated tool claims: "I ran", "I checked", and "the output shows" are only said when a tool call actually happened this session, and the honest form otherwise is "from memory" by @Natuworkguy
• Keep shell calls clean: a tool call contains only the command, never a comment or a narration by @Natuworkguy
• Every DONE line in a multi-part report carries the thing that proves it, quoted: the test output, the log line, the grep match by @Natuworkguy
• Cut the filler harder: no acknowledgement openers like "Perfect!" or "Got it!", no scare-quoting the user's words back at them, no ** bold in a chat reply by @Natuworkguy
• Sound more like a person across a whole session: vary the shape of replies from turn to turn, let paragraphs be lopsided, and repeat a word rather than reaching for a synonym by @Natuworkguy
• Write files to disk through the write tool and hand back the path, never a page pasted into chat as a code block by @Natuworkguy

Tip

Use Natuworkguy/flash-onyx-2.4:31b-cloudbase to try the cloud build, or flash-onyx-2.4:12b on consumer hardware.

Full Changelog: v0.4.3...v0.4.4

FLASH CLI v0.4.3

What's Changed

• Add voice mode: /voice on sets up a local Vosk speech model and a Piper voice, then listens, speaks the replies, and keeps listening between turns [BETA] by @Natuworkguy
• Say "interrupt" to cut a spoken reply off mid-sentence and go straight into the next message by @Natuworkguy
• Speak only the prose of a reply, skipping code blocks, tables, and URLs, and tell the model it is being heard rather than read by @Natuworkguy
• Stream update output and end a failed update with the last lines the command actually printed, instead of a bare exit code by @Natuworkguy
• Screenshot the page when asked to look something up on the web, rather than answering from a search snippet by @Natuworkguy
• Clear the screen on start and center the banner by @Natuworkguy
• Improve the README and polish the configuration docs by @Natuworkguy

Flash Onyx 2.3

• Add the flash-onyx-2.3 Modelfile by @Natuworkguy
• Gate the self-introduction behind an actual question about it, so a reply that follows a tool call opens with the work by @Natuworkguy
• Treat totals like "the only language" as claims about everything unchecked, and make "are you sure" mean check again rather than say yes again by @Natuworkguy
• Answer from the output already pulled instead of an impression of it by @Natuworkguy
• Add the research process it was missing: what would answer the question, how to query, breadth before depth, primary sources over summaries, and when to stop by @Natuworkguy
• Sound more like a person: vary sentence openings, mirror the register it is answering in, react to results that do not add up, and put bad news first and plain by @Natuworkguy

Tip

Onyx Cloud Base models are now live. Use Natuworkguy/flash-onyx-2.3:31b-cloudbase to try it out.
See this blog post for more info.

Model builds

• Add a push option to the build process by @Natuworkguy

Full Changelog: v0.3.3...v0.4.3

FLASH CLI v0.3.3

What's Changed

• Updater fixes by @Natuworkguy
• The agent can now interact with web pages by @Natuworkguy
• System prompt fixes related to the write tool by @Natuworkguy

No new models

Full Changelog: v0.3.2...v0.3.3

FLASH CLI v0.3.2

What's Changed

• Add MIT License to the project by @Natuworkguy
• Add licensing information to README and update build process for model licensing by @Natuworkguy
• Fix Ollama installation function to install scripts by @Natuworkguy
• Extract the Linux curl install path into install_ollama_curl() and reuse it on macOS when brew fails by @Natuworkguy
• Add PDF and Word document reading to the read tool by @Natuworkguy
• Update prompt to include current version in Flash update confirmation by @Natuworkguy
• Change thinking spinner speed to 5 by @Natuworkguy

Flash Onyx 2.2

• Add game development guidelines, emphasizing gameplay feel and mechanics by @Natuworkguy
• Adjust Onyx 2.2 model parameters by @Natuworkguy
• Add guidelines for effective argumentation by @Natuworkguy
• Add stock trends guidelines for market analysis by @Natuworkguy
• Add guidelines for understanding user requests by @Natuworkguy
• Add guidelines for naming files and selecting technology stacks by @Natuworkguy
• Add guidelines for quality standards and thoroughness by @Natuworkguy
• Clarify identity and output guidelines by @Natuworkguy
• Add guidance for screenshot tool delays to enhance animation capture by @Natuworkguy
• Improve Onyx 2.2 and make it better at manim by @Natuworkguy

Full Changelog: v0.3.1...v0.3.2

FLASH CLI v0.3.1

What's Changed

• Add more spinners by @Natuworkguy

Flash Onyx 2.1

• See "Onyx 2.1 in one line" in the link above

Full Changelog: v0.3.0...v0.3.1

FLASH CLI v0.3.0

What's Changed

• Installer fixes by @Natuworkguy
• Version-related fixes by @Natuworkguy
• Implement live output streaming for shell commands with timeout handling by @Natuworkguy
• Add retry mechanism for final response in chat function by @Natuworkguy
• Add system prompt build script by @Natuworkguy
• Add playwright support by @Natuworkguy

Flash Onyx 2

• Make Flash Onyx 2. See the link above for more info.

Full Changelog: 0.3.0...0.2.0

FLASH CLI v0.2.0

What's Changed

• Streaming replies. Model responses now render progressively, word by
  word, with a trailing cursor dot while the text is still coming in, no more waiting for the whole reply to print at once.
• /image path suggestions. Typing /image now suggests files and
  folders from disk as you type, filtered to supported image types
  (.png, .jpg, .jpeg, .webp, .gif, .bmp). Paths with spaces are
  automatically quoted so they parse correctly.
• Image-aware thinking states. Sending an image now cycles through a
  dedicated set of status phrases ("Examining the image", "Studying the
  image", ...) instead of the generic "Thinking" list.
• Rotating hint suggestions. An idle prompt now shows a rotating tip
  (drawn from flash/suggestions.json) that sweeps into view letter by
  letter and sweeps back out, cycling through example commands and
  prompts.

New Contributors

• @Pacsfury made their first contribution in #16

Full Changelog: https://github.com/Natuworkguy/Flash/commits/v0.2.0