Proposals for Power-User Features: Typography, Caching, and Session Memory Management

[7hebest]

🚀 Feature Proposals for Askimo - Power-User Experience

Hi everyone,

Following a productive exchange with the development team, I wanted to open up a few feature proposals to the community.
Having benchmarked several desktop clients, Askimo's native Skia-based architecture and Compose Multiplatform stack stand out as a highly efficient, responsive framework.

To further elevate the power-user experience, I would love to propose and discuss the following features.

---

🛠️ 1. Independent Typography Controls

While Askimo already supports global UI font selection, there is a clear benefit to introducing independent controls for code blocks.

Enabling users to set a dedicated monospace font (such as JetBrains Mono or Fira Code) for code displays, while keeping a standard sans-serif for the global UI, would significantly improve readability during deep-work sessions.

---

⚡ 2. API-Side Prompt Caching Support

For heavy, long-context sessions, cost and latency accumulate rapidly.

Integrating prompt caching by sending cache-control hints to providers that natively support it (such as Anthropic and OpenAI) would be a massive efficiency gain for API-based workflows.

---

🧠 3. Configurable Reasoning Effort Budgets

For models that support variable reasoning budgets (such as OpenAI's o3-mini), adding UI controls to toggle between Low, Medium, and High reasoning effort would allow users to:

• Balance API costs dynamically
• Tune execution time against depth of analysis
• Adapt the reasoning mode to the complexity of each task at hand

---

🎨 4. Native Theming & Layout Density (The Skia Approach)

Since Askimo renders natively via Skia and bypasses the Electron/DOM pipeline entirely, custom CSS injection is not technically applicable.

However, a native customization engine allowing users to configure:

• Core color palettes
• Font selections
• Layout spacing density

...would provide the perfect ergonomic solution for power users without introducing web wrapper bloat.

---

🧩 5. Advanced Session Memory Management (Context Drift & Decay Mitigation)

Askimo handles ultra-long sessions without client-side crashes, but instruction drift remains an LLM-level bottleneck over hundreds of turns.

Implementing background context management, such as:

• Configurable system prompt re-injection (e.g., every N turns)
• A rolling context window that condenses older history into an ephemeral "state summary" token

...would dramatically improve rule-adherence and maintain persona consistency during deep-work sessions.

---

💬 Closing Thoughts

I would love to hear thoughts from the community and the maintainers on these proposals and how they might fit into the upcoming roadmap.

[haiphucnguyen]

Hi @7hebest , Thanks so much for taking the time to write these up! These are fantastic ideas, and honestly, they align perfectly with where we want to take Askimo. I'm happy to confirm that all of these are officially on our roadmap, with specific attention going to the API-side caching and reasoning budgets for the very next release.

1. Independent Typography Controls: Add Separate, Deterministic Font Settings for UI Text and Code Rendering #435 (Askimo 1.4.9 or Askimo 1.4.10)
2. API-Side Prompt Caching Support: API-Side Prompt Caching Support - Anthropic cache_control Implementation #436 (Askimo 1.4.8)
3. Configurable Reasoning Effort Budgets: Configurable Reasoning Effort with Optional Per-Prompt Override #437 (Askimo 1.4.8)
4. Native Theming & Layout Density (The Skia Approach): Enhance Native Theming: Density Presets + Optional Advanced Theme Customization #438 (Askimo 1.4.9 or Askimo 1.4.10)
5. Advanced Session Memory Management (Context Drift & Decay Mitigation): Enhance Long-Session Memory: Configurable Re-injection Cadence + Optional Ephemeral State Summary #439 (Askimo 1.4.8 or Askimo 1.4.9)

Really appreciate the insight! Keep the feedback coming.

[haiphucnguyen]

Hi @7hebest,

I’m excited to share that Askimo 1.4.8 is officially live! A huge portion of this release was shaped by your excellent feedback, and I wanted to personally thank you for your contributions. We really value the perspective you’ve brought to the project.
Here’s a look at what’s included, driven by your insights:
• Custom Font Settings: You can now configure separate font preferences for UI text and code rendering.
• Prompt Caching: Integrated to help optimize performance and speed.
• Reasoning Effort: We’ve added configuration for reasoning effort. We’ve kept the implementation streamlined by letting the model handle the heavy lifting, as we found it delivers the best results for users without the need for manual budget management.
• Native Theming & Layout Density: Improved visuals and UI flexibility to match your preferences.

Also we have other improvements such as transparent the tokens usage per each conversation, drag-drop attachments support, also improve the overall performance of the application.

Please check it out when you have a moment. We’re always looking to refine the experience, so feel free to drop any further suggestions. Thanks again for your time and for helping make Askimo better!

[7hebest]

---

Hi @haiphucnguyen,

Thank you for the update. Delivering an architecture-heavy release like 1.4.8 within such a short timeframe is a remarkable testament to your team's engineering agility and the inherent efficiency of the Compose/Skia stack.

I am glad that the proposed ideas aligned so cleanly with Askimo’s overarching design philosophy. The integration of API-side prompt caching, independent typography controls, and native theming provides an immediate ergonomic and performance leap. Furthermore, keeping the reasoning effort implementation streamlined to handle the heavy lifting in the background is an elegant architectural choice that respects power-user intent without introducing UI bloat.

Askimo continues to solidify its position as a foundational tool for our team's deep-work and research workflows. We will be leveraging 1.4.8 - 1.4... heavily in our daily environments, and I look forward to sharing further technical observations and structural proposals as they naturally emerge from real-world application.

---

Congratulations to the team on an excellent release.

[richardchen874-sys]

Congrats on shipping 1.4.8 so quickly. On prompt caching, is the implementation capability-driven per provider, or currently tied to Anthropic/OpenAI-specific request fields? If Askimo already supports a generic OpenAI-compatible endpoint, a small capability layer for caching, reasoning fields, streaming, and tool calls could make additional providers much easier to test. I'm evaluating one API for access to multiple Chinese models and would be interested in comparing where compatibility actually breaks. Which edge case has been hardest so far?

[haiphucnguyen]

Hi @richardchen874-sys as long as the AI providers follow the open ai api spec, then the application will work like expected.

[richardchen874-sys]

Thanks, that is helpful. If Askimo already works with any provider that follows the OpenAI API spec, then it sounds like the main thing to validate is where provider-specific fields start to differ in real usage, especially around caching, reasoning parameters, streaming, and usage reporting.

I can prepare a small OpenAI-compatible test endpoint focused on official Chinese models, with a limited private-beta credit, so you can check basic compatibility inside Askimo without changing the integration architecture.

If you are open to testing it, you can reach me at richardchen874@gmail.com. I can send the base URL, model names, limits, and a few suggested prompts to verify chat, streaming, and usage behavior.

[haiphucnguyen]

We test askimo with popular providers such as lmstudio, ollama and many models there are some Chinese models such as Deepseek, Qwen. Depends on the providers, the behavior may be different. Support the prompt caching from the server side is out of Askimo's scope. You can download Askimo and trial by yourself.

[richardchen874-sys]

Thanks, that makes sense.

I agree that server-side prompt caching should stay outside Askimo’s scope if it depends on provider behavior. Askimo probably only needs to make the provider differences visible enough for users to understand what is working: streaming, usage fields, model naming, error format, and any caching-related response behavior.

I’ll try Askimo locally with a few OpenAI-compatible Chinese model endpoints and focus on where provider behavior differs in practice. If I find anything reproducible around DeepSeek / Qwen-style compatibility, I’ll share a concise report.

Is GitHub Discussions the best place to report provider-specific compatibility notes, or would you prefer issues for concrete repro cases?

[haiphucnguyen]

Thank you @richardchen874-sys . I am looking forward to your reports.

Is GitHub Discussions the best place to report provider-specific compatibility notes, or would you prefer issues for concrete repro cases?
Yes, feel free to open a new discussion. Or if you think there is an issue of Askimo then you can open the issue. I do not have the hard rule :)

[richardchen874-sys]

@haiphucnguyen Thanks, that works.

I’ll start with a new discussion if the findings are more general provider-compatibility notes, and use issues only for concrete reproducible problems in Askimo itself.

For the first pass, I’ll probably test a small set of OpenAI-compatible Chinese model endpoints around:

1. basic chat completion
2. streaming behavior
3. usage fields
4. model-name handling
5. structured output / JSON behavior
6. provider-specific error responses

If I find anything useful, I’ll keep the report concise and include the provider behavior, request shape, observed response, and whether it looks like an Askimo issue or just provider-specific behavior.