Roadmap
What we’re working on for each release, kept in step with how we plan on GitHub. Fixes to bugs, quality, and anything that gets in your way always come first; features follow. Plans change, so treat later releases as intentions, not promises.
Something missing or broken? Tell us on GitHub. See what already shipped in What's new.
Toskar 1.8.3
0 of 6 done- QualityAsked to run a destructive command, the model refuses and hands the command over insteadCore
- QualityFeature: check access from anywhere from outside, and show the relay's health to AdminsCore
- QualityInclude stderr and probe errors on Heimdall health eventsCore
- QualityName the requested asset and HTTP status when llama.cpp install failsCore
- QualityQuality: image recognition cases (objects, text, charts, video, and no-vision fallback)Core
- FixSlow first answer from a remote Toskar Pro: the model has to load firstCore
Toskar 1.9.0
0 of 14 done- QualityFeature: UI/UX audit of the app — look sharp, follow interaction best practicesCore
- QualityNative-speaker review: new Automations strings (#335)Core
- QualityReport any error or warning as a bug, from where it's shownCore
- QualityTypo tolerance: intent rules, memory commands, automation parsing, and search, tested in the quality suiteCore
- FeatureAdmins choose what each role and person can use: files, pictures, video, voice, and moreCore
- FeatureAdvanced image/video models with hardware-aware recommendations, guardrails, and minimum-requirements UXCore
- FeatureDocs: image and video model hardware requirements page (#529)Core
- FeatureFeature: Access from anywhere subscription (Toskar's relay, end-to-end encrypted)Core
- FeatureFeature: Auto-compact: name it, show it in the chat, and let people control itCore
- FeatureFeature: Push notifications in the Toskar apps (automations and more), through a Toskar push relayCore
- FeatureImage/video hardware checks: working-set estimates, graphics memory, verdicts (#529)Core
- FeatureImage/video models: update stable-diffusion.cpp and add the permissive lineup (#529)Core
- FeatureManage people from the CLI and API: create, list, change, delete, headlessCore
- FeatureMedia quality: real-hardware runs, visual judging, and provenance (#529)Core
Toskar 1.10.0
0 of 8 done- FeatureFeature: A Dashboard for admins and managers: usage, privacy, and health at a glanceCore
- FeatureFeature: Encrypt traffic between computers, and help encrypt data at restCore
- FeatureFeature: Private items — encrypted notes, lists, and more, managed from chatCore
- FeatureFeature: Rotate the weekly quality run through the model catalog, and publish a results matrixCore
- FeatureFeature: Sign in with a passkey (deferred from #206)Core
- FeatureIntegrations: Home Assistant, the *arr apps, and others, as triggers and actionsCore
- FeatureModels as a store: free, free with conditions, and premium models from providersCore
- FeaturePublished known limits and goal metrics: models × hardware, on the websiteCore
Later
Ideas and larger work we haven’t scheduled yet.
- FeatureAdd a dev container for UI and API workCore
- FeatureAdd request IDs that link HTTP logs to task and placement eventsCore
- FeatureAdd video.edit and video.extendCore
- FeatureArchitecture question: CPU node fallback execution strategy and zero-dep inferenceCore
- FeatureCase study: Toskar as a game AI backend (Dungeon Party), and a blog postCore
- FeatureCheck GPU acceleration and run the quality set on an NVIDIA laptop (RTX 3080)Core
- FeatureCheck in CI that the curl examples are valid shellCore
- FeatureConnect Google, Slack, and Notion with OAuthCore
- FeatureCore-to-Core access from anywhere, and buying it outside the app storesCore
- FeatureCredit bug reporters in release notes (GitHub, in-app, and store reviews)Core
- FeatureDocument how to build a runtime adapterCore
- FeatureFeature and case study: Toskar in CI — a GitHub Action and toskarctl askCore
- FeatureFeature: Check that links in answers actually openCore
- FeatureFeature: Kubernetes-native model deploymentCore
- FeatureFeature: Let the assistant know where and when you are, with your approvalCore
- FeatureFeature: Rent cloud GPU computers from the app and use them as nodesCore
- FeatureFeature: Report a bug from the app, or by asking the assistantCore
- FeatureFeature: Support regulated use (HIPAA, medical, legal) with a restricted-data mode, audit trail, and a compliance guideCore
- FeatureKeep only TOSKAR_APPS_TOKEN for the apps release dispatchCore
- FeatureLearn preferred tools, and keep context per projectCore
- FeatureMove the core build to Node 26Core
- FeaturePlugins: a framework and SDK for building Toskar extensionsCore
- FeatureReport a hardware compatibility result from your machineCore
- FeatureReport AMD GPU inference on LinuxCore
- FeatureReport Intel GPU detection and inferenceCore
- FeatureResearch: Distributed single-model inference across a Yggdrasil GridCore
- FeatureRuntime adapter: MLX on Apple SiliconCore
- FeatureRuntime adapter: vLLM on Linux with NVIDIA GPUsCore
- FeatureStream chat completions in the Python exampleCore
- FeatureTest llama.cpp release asset selection for every platformCore
- FeatureWalk the source install on a clean machine and fix the gapsCore