Product / Desktop · Early access beta
Move between OpenAI, Anthropic, Google, Azure, xAI, OpenRouter, and Ollama without surrendering your keys, your context, or your control — with local memory recall, kids-safe profiles, and parent oversight in the same desktop surface.
Product walkthrough
Actual captures from the current desktop beta. The point is evidence: how Valence handles conversations, local memory, parental controls, settings, and oversight inside one calm, inspectable surface.
These are real captures from the current desktop beta. Select a view, then open it full size for a closer read.
Capabilities
Valence keeps the working surface calm while giving power users the pieces that actually change outcomes: provider choice, tool access, persistent context, and inspectable system boundaries.
Switching
Start with Claude, get a second opinion from GPT-4o, finish with Gemini, and keep the same working thread. @mentions reroute the conversation instead of forcing restarts.
@claude, @gpt, and @geminiTooling
Connect stdio-based Model Context Protocol servers for filesystems, databases, APIs, and other task-specific tools when the model needs real leverage.
Parallel work
Run separate threads in parallel, keep provider selection per tab, and cross-reference the important parts when one workflow branches into another.
Profiles
Create workflow-specific system prompts and preferences for code review, writing, research, or any custom role without rebuilding the workspace every time.
Persistence
Conversations save locally as JSON so you can reopen exact context later, maintain your own archive, and keep the raw history on your own machine.
Posture
Valence does not proxy your traffic. API keys stay local and requests go directly to the providers you choose, including fully local flows through Ollama.
Surface
The desktop surface uses WebView2 for readable markdown, syntax highlighting, and a dense but calm composition suited to long sessions. Legacy ImGui mode remains available for overlay-style usage.
Memory
Use #remember and #recall to keep important knowledge available across sessions without leaving the device. Memory becomes part of the working surface instead of a hidden side system.
Evaluation
Send one prompt to 2–3 providers in parallel when you need a second opinion, a cost check, or a faster read on which model is right for the job.
Maintainability
SRP LEGO-style components, factory-driven provider adapters, and event-based coordination keep the system understandable as the product line expands.
Provider ecosystem
Use the best model for each job without leaving the product surface. Cloud, enterprise, and local options all live behind the same interaction model.
Fast generalist coverage for writing, coding, and multimodal tasks.
Cloud
Strong analysis, careful reasoning, and review-heavy workflows.
Cloud
Fast synthesis and broad model coverage for research and summarization.
Cloud
Fits controlled enterprise environments where deployment boundaries matter.
Enterprise
Another model family in the mix when you want a different read on the same problem.
Cloud
Useful when you want access to a broader model menu without leaving the Valence surface.
Gateway
The local-first path for private or air-gapped workflows that cannot depend on external calls.
Local
Technical foundation
The architecture favors explicit interfaces, service boundaries, and local persistence over hidden automation. That keeps the product easier to extend and easier to trust.
WebView layer
Modern HTML, CSS, and JavaScript deliver markdown, code rendering, and long-session usability through WebView2.
Orchestration
ChatWindow and the configuration flow coordinate lifecycle, navigation, and user-facing state without burying the logic.
Components
Header, message area, and input surfaces stay intentionally scoped so the interface can evolve without cascading rewrites.
Provider system
Provider adapters sit behind a shared interface and factory pattern so new model families can be added with clear seams.
Abilities & MCP
Built-in abilities, trigger systems, and MCP support give the model a practical action layer when the workflow calls for it.
Profiles
ProfileService manages custom personas, system prompts, and preferences so different workflows can keep their own operating discipline.
Persistence
ChatStorageService keeps full message metadata in JSON so sessions can be reopened, exported, and audited later.
Settings
SettingsFactory, validation, API key management, and provider configuration stay inspectable instead of disappearing into the background.
Theme
ThemeColors.ModernHive keeps the UI coherent across components so density and readability hold up as the surface expands.
Workflow fit
Valence is most useful when one model is not enough, context matters, and the operator wants direct control over where the work goes next.
Ask several models to work the same prompt and use the differences to sharpen the answer instead of picking one provider forever.
Use different model strengths for review, debugging, and documentation while keeping the same project thread and tool attachments visible.
Brainstorm with one provider, outline with another, refine with a third, then keep the history as part of the research trail.
When one explanation does not land, switch teaching styles instantly and keep the thread moving instead of starting over somewhere else.
Use expensive models for hard work, cheaper models for routine tasks, and local models when experimentation or privacy matters more than raw frontier performance.
Keep keys local, keep conversations local, and decide explicitly when cloud providers enter the loop.
Next step
Valence is built for people who want model choice, local control, and a working surface they can actually inspect. Purchase the early-access license, download the current build, or go deeper through the docs and roadmap.
