Releases
What's new in Chalie.
Chalie is in beta, and we ship in the open — rough edges and all. Each release earns a little more of your trust: better memory, sharper judgment, more understanding.
Latest
v1.0.0-beta
June 24, 2026🎨 A Brand-New Interface
- • Rebuilt From the Ground Up: Both the chat app and the Brain admin console have been rewritten as fast, modern apps with full dark and light theme support throughout.
- • Watch It Work, Step by Step: As Chalie works through a task, each step appears live as its own reply with its activity beneath it; completed steps fold into a tidy summary you can expand. Refreshing the page rebuilds the same view, so reviewing a past answer looks like watching it happen.
- • In Sync Everywhere: Open Chalie on more than one device or tab and your messages and its replies appear on all of them at once, so the conversation stays in lockstep.
- • Listen to a Whole Answer: The speak button now plays back an entire turn as one continuous narration instead of reading a single message at a time.
👁️ Chalie Can See
- • Image Understanding: Upload a photo, screenshot, or scan and ask about it — Chalie describes what's in it, reads text inside it, and remembers it. Pick which provider handles vision, and Chalie verifies a model can truly see before trusting it with images.
- • Find Pictures by What They Show: Every uploaded image gets a detailed description that's indexed for search, so you can later find a picture by its contents — objects, colours, text, layout — not just its filename.
- • Vision for Any Model: Even a model with no built-in vision can read an image, routed through your chosen vision provider, with a local text-in-image fallback when none is set.
🌐 Real Web Browsing
- • Hands-On, Not Just Reading: Chalie now drives a real browser — it can open a page, click buttons and links, fill and submit forms, scroll, go back, and work through multi-step flows, keeping the page open between steps.
- • Screenshots It Can Actually See: Chalie can capture a page and inspect the screenshot with its own vision, and the screenshot is saved as a document you can ask about again later.
🔗 Connect to Any MCP Server
- • Reach Out to Remote Tools: Chalie can now connect outward to remote MCP servers and use their tools, managed from a dedicated Inbound / Outbound section in the Brain — add, edit, test, enable, and disable servers in a few clicks.
- • Safe and Self-Maintaining: Remote tools obey the same per-channel Allow / Ask / Deny permissions as built-in ones, duplicate servers are merged automatically, and a background heartbeat keeps each server's tool list fresh and its status accurate.
🧠 Memory That Learns and Forgets
- • Memories Age Naturally: Each memory carries its own relevance clock — recent moments stay vivid, consolidated knowledge lasts far longer, and stale, weak, or superseded memories eventually fade away on their own, so recall stays fresh instead of drowning in old junk.
- • It Organizes Itself: Once enough has accumulated, related memories are grouped and rolled up into higher-level summaries automatically, with no manual curation — and when Chalie learns something that contradicts an old fact, the old one is retired cleanly so it stops resurfacing.
- • Knows Where It Came From: Memories are weighted by their source — your own words carry the most weight, while background and automated activity is tuned down — so what you tell Chalie directly shapes what it remembers.
- • Picks Up Where You Left Off: A new conversation opens with a quiet, curated recollection of the most relevant facts and recent moments, and you can pin a reply to keep it verbatim forever — never rewritten, never decayed.
🔬 Thinks While You Rest
- • Proactive Research: A new background mind quietly researches topics it expects to matter to you, surfacing what it finds when it becomes relevant — and you can review exactly what it explored in a new Auto Research view in the Brain.
- • Learns Facts on Its Own: A background reflection step reads your recent conversations and proposes durable facts, updates, and corrections, so Chalie's understanding grows passively without you ever saying 'remember this'.
🃏 Rich Cards That Stick Around
- • Durable Tool Cards: The visual cards Chalie shows — weather, timers, lists, articles, and more — are now saved properly, so they stay intact when you scroll back through your history instead of collapsing into plain text, and they show up correctly even when produced partway through a longer answer.
- • Always Live, Never Stale: Chalie no longer answers time-sensitive questions — weather, news, prices, schedules — from memory; it checks a live source that turn.
🛡️ Smarter, Safer Permissions
- • One Clear Set of Rules: Permissions are now a single, readable list across three contexts — you in chat, Chalie's own background work, and other agents — each action set to Allow, Ask, or Deny, with header buttons to set a whole context at once.
- • Shell Commands Judged by What They Do: A command is classified by what it actually does, not by a label the model picks, so a destructive command can't wave itself through by calling itself harmless. Read-only commands run freely; the rest ask first.
- • Safe by Default: Sensitive actions are pre-set to deny in background and external contexts where no human is watching, anything unknown defaults to asking first, and your existing choices carry over automatically on upgrade.
💾 Full Backup & Restore
- • Whole-Instance Import/Export: An Import / Export section in the Brain snapshots your entire Chalie — memory, documents, skills, and settings — into a single file you can save anywhere, optionally encrypted with an AES-256 password.
- • Safe to Restore: Restoring does a careful, checksum-verified wipe-and-replace that rolls back automatically if anything goes wrong mid-swap, so a half-finished restore can never corrupt a running instance.
🔌 Easier Provider Setup
- • Guided Setup Wizard: Adding an AI provider is now a step-by-step flow — pick from a curated catalog of popular providers, and Chalie pre-fills the connection details, fetches the real model list to choose from, and only asks for an API key when one is actually needed.
- • Pick a Provider for Background Work: A Delegate setting lets you choose which provider handles Chalie's background research and web browsing, separate from your main chat provider.
📍 Location Awareness
- • Named Places: Save the places that matter — home, office, the gym — and refer to them by name, with conversations tagged by where they happened.
- • Departure Reminders: Ask Chalie to remind you of something when it's time to leave a place, with a sensible warning buffer before you need to go.
⚙️ Reliability Under the Hood
- • Long Conversations Don't Break: When a conversation outgrows the model's limits, Chalie distils older history into a running summary and carries it forward, so who you are and what you're in the middle of survives instead of being silently dropped — and it does this work while you're away so your next message comes back faster.
- • Failures Tell the Truth: Every built-in tool now reports a real error when something goes wrong instead of returning a plausible-but-wrong answer, so Chalie can self-correct or honestly tell you it couldn't do something.
- • Self-Healing Connection: Chalie detects when a live connection has quietly gone dead — even when the network never signals a drop — and reconnects on its own, so the chat keeps working without a manual refresh.
- • Smoother and Steadier: Faster, more reliable file search; a firm limit on how long any single model request can hang so a stuck call no longer freezes a turn; and a quicker, more dependable startup.
v0.8.0
May 29, 2026🧠 Skills That Grow With You
- • Self-Refining Skill Library: Chalie now ships with dozens of ready-made Skills — reusable playbooks for everything from market sizing and contract review to meal planning and system health checks. They're available the moment you need them, with no setup.
- • Build Your Own Skills: Ask Chalie to create, edit, or remove a skill in plain language, then browse them all in the new Skills tab of the Brain dashboard.
- • Proactive Suggestions: When Chalie notices it's repeating a complex set of steps, it can quietly suggest turning that into a reusable skill — so common tasks get faster over time.
- • Cleaner Permissions: The policy controls in your dashboard are now generated automatically from Chalie's own abilities, so what you see always matches what Chalie can actually do.
🛡️ Safe Files & Commands
- • Search Your Files: A new search tool finds files and text across your system quickly and safely, returning matching snippets with context.
- • Download From the Web: Chalie can now fetch a file from a URL and save it locally, with built-in safety checks on every address.
- • Write Files Safely: Chalie can create and update files, but always reads an existing file first before overwriting it — so nothing is changed blindly.
- • Run Shell Commands With Guardrails: Chalie can now run shell commands when you need it to. Every command is classified by risk, secrets are stripped from its environment, and genuinely destructive commands (like wiping a folder) are blocked outright. You stay in control with per-command confirmations.
- • Adjust File Permissions: A dedicated tool can change file permissions, showing you the before-and-after every time.
⏹️ Full Control of Chalie's Work
- • Stop Anytime: A new stop button lets you halt Chalie mid-task. There's no longer a fixed limit on how long Chalie can work on something — it runs until the job is done or you stop it.
- • Redirect On the Fly: Send a new message while Chalie is working and it will adjust course without you having to stop and start over.
- • See and Steer Background Helpers: When Chalie spins up background helpers for bigger jobs, you can now list them, talk to them directly, or stop them individually.
📍 Location Awareness
- • Named Places: Save the places that matter — home, office, the gym — and refer to them by name.
- • Departure Reminders: Ask Chalie to remind you of something when you leave a place, not just at a set time.
- • Location-Aware Memory: Chalie can tag conversations with where they happened and recall memories filtered by place, learning the patterns tied to where you are.
🔐 Vault Backup & Recovery
- • Your encrypted vault is now automatically backed up, so a password-verified recovery can restore access if something goes wrong. Recovery tries every backup and safely skips a damaged one, so a single bad file can never lock you out.
🌐 Network & Smart Devices
- • Ubiquiti Network Control (Early Alpha): An early alpha capability that connects your UniFi network and lets Chalie manage clients, devices, firewall rules, DNS, and VLANs — through either your local controller or UniFi Cloud, set up from a dedicated tab in the dashboard. Expect rough edges as we refine it.
✨ Interface & Experience
- • Hover for Details: Each step in Chalie's action trail now shows a plain-language summary when you hover over it, so you can see exactly what it did at each stage.
- • Token Usage at a Glance: A new usage table in the Brain dashboard sits below the chart with full time-slot coverage, and the day view now shows only today's hours.
- • Tables in Chat: Chalie can now present information as proper formatted tables right in the conversation.
- • Smoother Typing: Pressing Enter now adds a new line — use the send button or Shift+Enter to send — so multi-line messages are effortless.
- • Polish: Provider API-key links are clickable, OpenAI-compatible providers list their models correctly, confirmation pop-ups now match Chalie's theme, and light-theme contrast has been improved across dialogs and scrollbars.
🐛 Reliability & Fixes
- • Better Document Handling: Uploads now wait long enough for large files, report success only when a document is truly ready, and read text out of images automatically.
- • Clearer Tool Results: Code evaluation always returns a meaningful result instead of finishing silently, and when a tool hits a problem the reason is now surfaced rather than quietly dropped.
- • Sharper Memory: Behavioural patterns Chalie has seen before are reinforced when they recur and now show up properly in memory recall.
⚙️ Under the Hood
- • Modern Dependency Management: Chalie now uses pyproject.toml and uv, with voice and browser components installed only when you need them — for a lighter, faster setup.
- • Leaner Codebase: Removed several unused subsystems — the legacy interface SDK, the PWA wrapper, and a dormant self-introspection service — trimming thousands of lines for a more focused, reliable app.
- • Default Port Changed: Chalie now runs on port 31025 by default.
v0.7.1
May 22, 2026Improved AI Provider Selection
- • See All Available Models: If you use AI services that are compatible with OpenAI, you will now see a complete list of specific AI models available within the Brain UI. This makes it easier to choose the exact model that best fits your task.
Stability and Setup Fixes
- • Cleaner Setup Display: We've resolved a minor issue that could cause display problems with special characters during installation, ensuring a seamless setup experience.
v0.7.0
May 19, 2026🧠 Smarter Interactions & Abilities
- • Conversation Review: A new capability lets you easily review past transcripts and dive deep into previous interactions for better context.
- • External Integrations: We’ve introduced new architecture for seamless communication with external agents and tools, allowing Chalie to work within a broader ecosystem.
- • World State Clarity: The "What Chalie Sees" section in your World State tab is now available with a Formatted/Raw view switcher, giving you total visibility into Chalie's current knowledge base.
✉️ Powerful Communication Hub
- • Advanced Inbox Control: You can now manage your emails directly through Chalie: delete messages, mark them as read, designate them as important, or move them straight to spam.
- • Full Email Workflow: Beyond reading, you can now send, reply to, and forward emails using your standard SMTP services.
- • Improved Reliability: We’ve polished how Chalie handles incoming messages, ensuring that message bodies are never truncated and that spam detection is more accurate based on your server settings.
🏠 Automation & Organization
- • Home Automation Support: Gain the ability to connect to and control your smart home devices directly through Chalie, including six new automation actions.
- • Richer Context: Your calendar and contact information are now displayed in richer, more useful card formats within the interface, providing quick visual summaries.
- • Reliable Contacts: Contacts are now stored using full profiles, and synchronization with your contacts service is more stable than ever.
✨ Experience & Performance Upgrades
- • Sleeker Interface: Enjoy a complete redesign of the Chalie brain dashboard, featuring modular panels and a refined visual look.
- • Voice Improvements: Transcription accuracy has been significantly improved, and we've fixed issues related to stuttering and filler words, making voice commands feel more natural.
- • UI Refinements: The Task Strip has been moved to a convenient right-side slideout drawer, and we've implemented theme-aware components for a consistent look whether you use light or dark mode.
- • Internationalization: Localization support has been streamlined, improving consistency for all users around the globe.
v0.6.2
May 12, 2026🎤 Voice Subsystem Rebuild
- • Rebuilt voice on kokoro-onnx (TTS) + moonshine-onnx (STT) — both ONNX, fully on-device.
- • espeak-ng ships as a wheel dependency via espeakng-loader — no system package needed.
- • Markdown-to-speech preprocessing: markdown-it-py + nh3 chain, pysbd sentence segmentation, num2words ordinal expansion.
- • Per-sentence NDJSON streaming for responsive playback.
- • Long-audio STT: chunked transcription supports up to 10 minutes total in 60-second windows.
- • URLs read as host only: http://google.com/123 becomes 'google dot com'.
- • TTS error sentinel is now disjoint from the done sentinel — fixes a stuck-spinner bug on synthesis failure.
⚙️ Installer
- • Added Alpine (apk), Arch/Manjaro (pacman), and openSUSE (zypper) system-dependency branches.
- • Better diagnostics when voice-model downloads silently fail during install.
🧹 Dead-Code Removal (rc-0.7.0)
- • Ripped: MemoryStore Redis-compat shim (15 methods), ACT output queue, ability pre_dispatch/post_dispatch hooks.
- • Removed: predict_multi_label, VaultService.change_password, get_lists_for_prompt, ConfigService.load_text + PROMPTS_DIR, news CATEGORIES/search_sources.
✨ UX
- • Brain provider form: model list auto-fetches on platform, host, or API-key change — Refresh button removed.
🐛 Fixes
- • Schedule recall restored.
- • Windows blank-UI fix: JS MIME types registered at process start, overriding stale registry values.
- • chalie update now correctly passes the release tag through to the installer.
v0.6.1
May 10, 2026🚀 Setup & Providers
- • Smarter Model Selection: When adding new AI services, the list of available models will now load automatically. You no longer need to manually refresh the list.
- • Instant Activation: When you add a new AI provider, Chalie will automatically switch to it, so you can start chatting right away.
- • Improved Stability: We’ve improved how Chalie refreshes and manages connected providers to ensure a smoother experience after setup.
🎤 Voice Experience
- • More Natural Speech: The text-to-speech engine has been updated to better understand punctuation, resulting in more naturally paced and articulate responses.
- • Improved Loading: Mic and speaker icons are now hidden until Chalie is fully ready, preventing confusion during startup.
- • Voice Fixes: We’ve resolved several underlying issues related to voice loading and transcription, ensuring the assistant is ready to go reliably on first use.
✨ Interface & Stability
- • Better Visibility: We fixed a minor issue where certain labels in the chat history were invisible when using the light theme.
- • Connection Refresh: The system now reliably refreshes the list of available models against the latest service information.
- • Cleaner Start: We’ve streamlined internal processes to ensure a quicker and cleaner startup experience.
v0.6.0
May 8, 2026🎨 Rich-Media Chat Cards
- • Weather, lists, timers, scheduler, and article results now render as proper visual cards inline in chat — not just text. Weather has an ambient-sky background that shifts with the conditions; article cards carry a real image below the synthesis; lists keep their checkboxes between turns.
- • Built-in countdown timer — ask Chalie to set a timer and you get an ephemeral card with pause, resume, and an alarm when it ends. The timer survives pause/resume cleanly and counts down to the second.
- • Smarter image picking — search and news results now shortlist candidate images with descriptive captions so the model can choose the one that matches the story, instead of grabbing whatever happened to be first. When a source has no thumbnail, Chalie falls back to its og:image automatically.
- • Same-day scheduler view — when you ask about today's events, the card lists everything for the day and drops the redundant confirm button.
✨ Interface Refresh
- • Light and dark themes with a persistent toggle — Chalie remembers which one you picked across sessions.
- • Full-bleed assistant turns — Chalie's responses now flow edge-to-edge without bubble chrome, with a glyph + timestamp header and 24px action buttons. Cleaner read, more breathing room.
- • User turns simplified — bold text, no metadata row, an 80% white bubble in light mode that disappears entirely in dark.
- • Subtle living details — a faint stripe accent, soft bloom on the active turn, hover lift on action rows. Restraint over flash.
- • Themed audio player and ACT spinner — the inline tool-narration spinner now adapts to whichever theme you're in.
🤖 Subagent Rework
- • `goal_pursuit` is now `subagent` — a single tool with a type-discriminated schema (web_surfer, researcher, etc.). The model picks the right specialist by name, and the wait/cap behaviour is tuned per-type (web_surfer can run synchronously for up to 30 minutes).
- • ACT timeout removed — long-running subagents no longer get killed mid-task by an arbitrary cap. The dispatcher waits as long as the subagent type allows.
- • Async subagent results now reliably reach the user — three silent-death paths in the delivery pipeline have been eliminated, including a race on the parent processor lifecycle.
- • Schema collision fixed — the subagent's `type` parameter was colliding with the dispatcher's reserved action-envelope key, causing 'Unknown action type' errors. Renamed to `agent_type`.
🧠 Memory & Routing
- • Recall is now single-call-per-topic — the model is guided to make one focused recall query per topic instead of building intra-query splitting into the engine. Plus narrow-retry guidance when the first call comes back thin.
- • Schedule search via the schedule tool, not memory — recall no longer fans out into scheduled items. The model calls the schedule tool directly, with a new semantic `search` action backed by vector similarity.
- • Document and schedule recall now route through the real tool dispatch path — eliminating duplicated query logic and synthetic record-and-render workarounds.
- • Concept canonicalisation hardened — aliases like 'birthday' vs 'birth_date' now resolve reliably because the lookup table embeds aliases as their own rows (27 → 365 rows) instead of relying on cosine similarity to bridge the gap.
- • find_tools hoisted to the front of the always-available list — when the model is choosing what to call first, tool discovery is now its first option, not its last.
- • Search routing supplemented with DuckDuckGo on weak signals — when the router is uncertain, a lightweight DDG pass broadens the result set instead of returning thin output.
🗜️ Append-Only Compaction
- • Compaction has been rewritten as an append-only flow over `tool_calls` with a continuity prompt and a clean subagent split. No more rewriting history — past turns stay verbatim, and the model picks up exactly where it left off.
- • `max_tokens` and `compact_at` are now persisted per-provider — no on-the-fly threshold math, no 'why did it compact now' confusion.
- • Provider payload measurement fixed — compaction now measures the actual provider payload, not just the user body, so the threshold reflects what the model will actually receive.
- • Proactive overflow no longer loops — a one-shot guard prevents runaway compaction when the handler returns False.
🎙️ Voice
- • Parallel TTS — three independent Kokoro instances now synthesise in parallel, dropping wait times on long messages from seconds to fractions. Memory cost ~600MB; latency win is dramatic.
- • Markdown bleed fixed — `*example*` is no longer pronounced 'asterisk example asterisk'. Bold, italic, code, links, and bare URLs are stripped before synthesis. Identifiers like `module_name` and expressions like `2 * 3` survive intact.
- • Silent truncation removed — phoneme-cap overflows now raise loudly, and text is pre-chunked along sentence and clause boundaries so the cap is never hit. Audio no longer goes missing without a log line.
- • Synchronous NDJSON streaming — fixes a 50% chunk-loss bug where TTS frames were being dropped mid-stream.
- • Kokoro fp16 patch — works around a kokoro-onnx 0.5.0 dtype bug that broke synthesis on the HF fp16 schema.
- • Player keyboard shortcuts no longer hijack typing — chat input has full keyboard focus while the player is open.
📎 Reliability & Polish
- • File uploads block the LLM until processing completes — no more silent skips, no more 'apologise and move on' when the model races past an attached file.
- • Path traversal guard on data file endpoints — get/put/delete now resolve real paths before reading.
- • Read tool preserves dates and URLs — trafilatura now extracts with `include_links=True`, so date metadata that lives next to anchor links survives the boilerplate filter.
- • Privacy delete-all reaches all live tables — three dead tables removed from the wipe path, replaced with the actual current set.
- • Markdown is now forbidden inside allowed HTML tags — system prompt explicitly rules out the mixed-syntax outputs that were sometimes leaking through.
- • Brain dashboard error log — surface the latest backend errors directly in the dashboard via `/system/observability/errors`, instead of hunting through log files.
⚙️ Under the Hood
- • DMN folded into the subconscious worker as step 5 — the background-LLM queue is gone, and DMN ticks now ride the same idle-gated pipeline as everything else.
- • Provider configuration flattened — single global LLM provider, with `max_tokens`/`compact_at` persisted instead of recomputed on every call. Single backfill function shared between boot and provider creation.
- • Canonical paths — `paths.py` now defines sibling `/data` and `/resources` directories; environment-variable and CLI-flag overrides have been dropped.
- • Dashboard gateway moved into the backend, dead `app.py` ripped out.
- • Sonar pass — seven sweeps clearing 51 issues across PR-1698 and follow-ups: unused locals/params, commented-out code, redundant exception classes, duplicate string literals, JS modernisation, form-label associations, log-injection sanitiser, explicit Flask methods.
- • Substantial dead-code removal — DataGraphService lost 8 unused public methods, DocumentService lost a dead `search_by_metadata`, the LLM service dropped an unused `stream=` parameter, and several other helpers with zero callers were retired.
v0.5.1
April 30, 2026🛠️ Bug Fixes
- • Docker image now ships with classifier heads — Every Docker user was running with the deliberation gate stuck at `low` and mode_detector dead because the published image was missing pre-trained ONNX classifier files. The Dockerfile now installs Chalie via the same `install.sh` path native users use, so the image is identical to a fresh local install.
- • Native tool calling fixed — Some cloud models were encoding tool calls as XML tags inside their text response instead of using their provider's structured tool-call channel. Tools were never executing, but the model would write convincing prose claiming they had. The system prompt has been reworded so the model is told the exact response field where its prose lands per provider, and the inline XML fallback that masked the problem on some models has been removed.
- • No more silent tool-call rescues — All four LLM provider adapters (Ollama, Anthropic, OpenAI, Gemini) now read tool calls exclusively from the structured response field. If a model fails to populate it, the turn fails loud instead of being silently rescued by content-side regex.
🧹 Cleanup
- • Dockerfile slimmed — Build args now match `install.sh` flag names verbatim (`--disable-voice`, `--branch=NAME`); runtime args pass straight through to `run.sh` (`--port`, `--host`, `--no-voice`). No invented umbrella flags.
- • Useless tests deleted — A handful of tests that only checked for specific substrings inside Python string constants have been removed. They were locking variable values, not testing behaviour.
v0.5.0
April 29, 2026🧠 Sharper Cognition
- • Subconscious Worker: A single 5-minute idle-gated tick now handles all of Chalie's latent cognition — memory consolidation, decay, pattern matching, and self-summary. Replaces four separate background daemons with one cleaner pipeline.
- • Smarter Pattern Matching: Chalie now identifies behavioural patterns in a single LLM pass, with confidence tracking that strengthens over time and gracefully decays unused observations.
- • Live Mode Gate: Chalie now dynamically adjusts its prompting style based on what you're asking — math, code, planning, conversation — without you having to flip any switches.
- • Better Memory Recall: Recall now ranks results by true semantic similarity, not just how often something has been retrieved. Fewer noisy hits, more relevant results.
✨ Cleaner Chat Experience
- • Chrome-less Thinking Indicator: While Chalie thinks, you see a single elegant blinking logo with inline tool narration — no more chunky bubbles. When the answer arrives, the whole indicator collapses into the response.
- • Auto-Linked URLs: Drop any URL in a message and it becomes clickable on render. Cleaner output from the model, no more manually-formatted link tags.
- • Faster, More Visible Performance: Real-time tool start/end events plus chat-timing instrumentation surface actual response latencies in the UI.
⚡ Lighter Runtime
- • Goals System Removed: Chalie no longer maintains a separate Goals system — natural conversation surfaces the same intent with less ceremony.
- • Adaptive Layer Removed: Style metrics and adaptive directives are gone — Chalie follows its core personality directly, no extra layer in between.
- • Ambient Awareness Consolidated: AmbientInference, SituationModel, and PlaceLearning have been replaced by a unified WorldState with typed signal absorption.
- • Decay Daemon Retired: Memory decay now runs as part of the unified subconscious tick, not as a separate background thread.
- • Streamlined Provider Setup: A single global LLM provider replaces the per-job provider assignments. Onboarding is now login + provider, with advanced multi-model setup available in Brain.
🛠️ Reliability & Polish
- • Hardened HTML Pipeline: Every assistant response is now sanitised through industry-standard tooling before it reaches the chat — strips anything outside Chalie's allowlisted tags. The frontend trusts the backend, no double-checking.
- • Ollama Cloud Recovery: When a request exceeds the cloud edge proxy's body-size limit, Chalie now triggers an emergency compaction and retries cleanly instead of failing the turn.
- • Tokenizer Quirks Cleaned: Qwen-style escape sentinels in tool arguments are now stripped at the dispatcher chokepoint — affects all tools without per-ability fixes.
- • Safer Inference: NaN/Inf guards on classifier paths, real-asset feature tests, sha256-pinned encoder. Bad inputs return clean 'no answer' rather than degenerate scores.
- • Scheduling Fix: Duplicate reminders are now blocked at the database level, even when the model invokes schedule() twice in the same turn.
- • Cleaner Codebase: Major code health pass — net -770 lines on the markup pipeline alone; 14 abilities ported to a unified ABC; zero static-analysis suppressions added across the entire release cycle.
v0.4.0
April 23, 2026🧠 Smarter Intelligence & Memory
- • Deeper Understanding: Chalie now uses a more advanced system for thinking through problems. It can explore different approaches before settling on the best answer, leading to more thoughtful and robust responses.
- • Enhanced Personalization: Your personality profile is now driven by a new, customizable system. You can fine-tune Chalie's tone and style using a 5-slider interface.
- • Improved Memory: We've refined Chalie's memory system. It's now better at identifying facts, linking new knowledge to what it already knows, and handling conflicting information more gracefully.
- • Better Learning: When Chalie learns from your conversations, the process is cleaner and more structured, ensuring new facts are stored precisely where they belong.
✨ Interface & Experience Polish
- • Voice Control Upgrade: We've completely rebuilt the voice interaction experience. Chalie now features a cleaner, inline microphone button and a more polished playback experience.
- • Faster Interactions: We've optimized several background processes, meaning Chalie can retrieve and summarize relevant information faster during your chats.
- • New Ways to Use Chalie: We've introduced a compatible platform for third-party AI services (like Groq or MiniMax), giving you more options for how Chalie runs.
- • Cleaner Setup: The onboarding process has been streamlined to focus purely on setting up your accounts and connection preferences, leaving the advanced setup to the main application.
🛠️ Stability & Performance Improvements
- • Rock-Solid Connections: We've improved how Chalie handles connections to different services, especially with Ollama, ensuring better detection and fewer unexpected pauses.
- • Reliable Summaries: The way Chalie builds summaries of your interactions is more robust, ensuring that key context is never lost during long conversations.
- • Better Performance: Behind the scenes, we've made multiple optimizations to how Chalie processes large documents and information chunks, resulting in snappier results, especially with lengthy inputs.
- • Fixed Bugs: Various bugs have been squashed, including issues with how tools and knowledge are recognized, and fixes to ensure metric tracking is accurate even when things go wrong.
v0.3.3
April 18, 2026🚀 Enhanced AI Options
- • You can now connect Chalie to a wider range of external AI services. If you prefer using an AI platform other than the default, you have more options available to connect and power your assistant.
✨ Smoother Conversations
- • We've improved how Chalie processes responses. You should now see cleaner, more focused answers, as we've removed any internal processing notes that used to appear in the chat.
v0.3.2
April 14, 2026✨ Smarter Memory & Recall
- • Deeper Context: Chalie now uses a much more advanced system to recall relevant information from your history, allowing it to give contextually accurate answers across longer conversations.
- • Personalized Learning: Your ongoing conversations now influence Chalie's understanding of your needs in real-time, making its responses feel more tailored to you over time.
- • Improved Conversation Flow: We've cleaned up how Chalie manages its long-term knowledge, resulting in faster, more stable interactions and fewer confusing breaks in topic.
🛠️ Performance and Reliability Boosts
- • Faster Experience: We've optimized several core processes, which means Chalie should feel snappier and more responsive during complex tasks.
- • Stricter Security: Improved handling for sessions ensures Chalie remains secure, especially when switching between different contexts.
- • Better Updates: The process for updating Chalie's own documentation and knowledge base is now fully automated, ensuring Chalie is always current with its latest information.
📰 Knowledge & Tool Improvements
- • Simplified News: Our news tool is now easier to use, requiring less input from you while still delivering relevant stories based on the topic and region.
- • Smarter Tool Use: Chalie has an improved way of figuring out which of its various skills or tools to use for a task, making its responses more accurate when it needs to perform an action.
🐛 Bug Fixes
- • Stable Sessions: Fixed issues that could sometimes cause Chalie to get stuck or lose your place during long interactions.
- • Data Consistency: Corrected background processes to ensure that any information Chalie stores or accesses is accurate and up-to-date.
- • Application Resilience: Improved how Chalie handles restarts and unexpected events, ensuring a quicker and more stable return to normal operation.
v0.3.1
April 4, 2026New
- • Contradiction-at-creation — replaces the uncertainty engine; contradictions are detected and resolved at write-time instead of accumulating stale records
- • Durable auth sessions — sessions survive restarts, chat history loads from SQLite instead of MemoryStore
- • Cross-thread chat history — conversation history persists across threads on page load
- • User timezone support — detect, persist, and respect your local timezone
- • Automated release announcements via n8n + Ollama + Telegram
Improved
- • Document search — simplified from 3-stage cascade to flat 2-signal (chunk semantic + FTS5); vector-only scores boosted when FTS returns nothing
- • Digest worker — extracted singletons and post-exchange hooks into focused modules
- • Schema cleanup — compressed migrations, added composite indexes, fixed reliability column bug
- • List skill — tighter tool schema for weaker LLMs, defensive JSON-string array parsing
- • Build log pipeline — moved LLM summarisation to n8n/Ollama, GH Actions is now a dumb webhook trigger
- • Webhook security — opaque URL tokens + X-Webhook-Secret header
Removed
- • Focus skill + FocusSessionService
- • Reflect skill + ReflectAction
- • Uncertainty engine (replaced by contradiction-at-creation)
Housekeeping
- • Resolved all ruff + vulture findings (230 errors → 0)
- • Removed stale template placeholders from frontal-cortex prompts
- • README rewritten for technical substance
v0.3.0
April 2, 2026Email, Calendar & Contacts
- • Connect your email, calendar, and contacts — Gmail, Outlook, iCloud, Fastmail, or any standards-compliant provider
- • Read, search, reply to, and send emails conversationally
- • Browse calendar events, detect conflicts, and find free slots
- • "Email Sarah" just works — Chalie resolves the right person from your contacts
Proactive Intelligence
- • Morning brief — daily digest of your calendar and inbox highlights
- • Meeting prep — attendee context and notes surfaced before your meetings
- • Conflict alerts — overlapping events flagged with reschedule suggestions
- • Departure nudges, draft reply suggestions, and welcome-back catch-up when you return
- • Quiet windows — proactive prompts suppressed during off-hours and focus time
Smarter Goals
- • Goals now emerge from patterns in your experiences — not keyword matching
- • Casual intent like "I was thinking about..." now creates goals naturally
- • Chalie learns from outcomes — improving how it approaches tasks over time
New Tools
- • Browser — browse the web, take screenshots, interact with pages
- • News — stay current with RSS feeds and world awareness
- • Rich render — visual output with tables, charts, and cards
Security
- • AES-256-GCM envelope encryption — all stored credentials are now encrypted at rest
Stability
- • 2,997 passing tests — 360 commits across 375 files
- • Removed ~8,000 lines of dead code — leaner, faster, more reliable
v0.2.1
March 23, 2026Stability & Production Fixes
- • Encryption key now survives Docker rebuilds — no more lost API keys after container updates
- • Fixed database migration crash when upgrading from v0.1.0
- • Short messages like "yes" and "ok" no longer trigger false topic switches
Smarter Memory
- • Conversation summaries now trigger reliably on long threads — no more lost context
- • Internal system events no longer pollute your memory — only real conversations are remembered
Better Goals
- • Duplicate goals are now caught before they form — smarter matching prevents redundancy
- • Proactive suggestions feel more natural — tuned when and how Chalie reaches out
- • Stalled background tasks now resolve cleanly instead of getting stuck
Dashboard
- • New token analytics panel — see AI usage costs and response times at a glance
- • Automated pipeline health checks — Chalie monitors its own systems
- • Camera and image providers auto-detected — no manual vision model setup
Testing
- • 2,875 passing tests — zero regressions from v0.2.0
v0.2.0
March 22, 2026Goals
- • Chalie now tracks what you're working on — goals form from your conversations and strengthen over time
- • Ask "what are my goals?" to see what Chalie has picked up
- • Confirm, complete, dismiss, or mute goals naturally
- • Chalie proactively helps with mature goals — checks in naturally, not with robotic templates
- • Related goals are automatically linked into hierarchies
- • Ask "tell me about my goals" for a narrative of how they've evolved
Conversations
- • Single, smarter response path — faster replies, better skill and tool selection
- • Chalie remembers more of your conversation — smart summarisation keeps context across long threads
- • Better topic detection — knows when you've changed subjects
Apps
- • External apps can now pair with Chalie (Gmail, Calendar, Git, Stocks, News, and more)
- • Manage connected apps from the chat interface
- • Apps expose tools that Chalie can use autonomously
Voice
- • New voice engine — faster, lighter, runs entirely on your hardware
Documents & Images
- • Take photos or attach images directly in chat
- • Better document search and organisation
Interface
- • Cleaner, modular chat UI
- • Token usage stats in Brain dashboard
- • In-app updates
Stability
- • 2,640 passing tests
- • Removed ~6,500 lines of dead code
- • Fixed dozens of crash-causing bugs
v0.1.0
March 3, 2026First Alpha
- • First official alpha release
- • One-line installer for macOS and Linux
- • Docker image on DockerHub
- • Persistent memory — Chalie remembers you across sessions
- • Multi-provider AI support (Ollama, OpenAI, Anthropic, Gemini)
- • Local-first — all data stays on your machine
- • Background thinking, task execution, and document intelligence
- • Voice input and output
- • Brain dashboard for observability