Getting Started
AI Calculations
Hissab can use an LLM to translate natural language into Hissab expressions, evaluate those expressions with the calculator engine, and answer from the results.
Bring your own provider key
Open Settings and connect a provider by saving an API key. The app has provider entries for OpenAI, Anthropic, Google Gemini, DeepSeek, and a custom endpoint.
For a custom endpoint, choose the SDK / protocol shape:
- OpenAI-compatible
- Anthropic Messages API
- Gemini GenAI API
Provider keys are used directly from the browser. Only add keys you are comfortable using from this app and browser profile.
Enable models
Each provider has a model list. Check the models you want available, then pin one enabled model as the default — the default is what the composer uses for new questions.
Built-in providers include static fallback model lists. For OpenAI, Anthropic, Gemini, and DeepSeek, the app can refresh available models from the provider when an API key is present. Custom endpoints use the model IDs you add manually.
How AI calculation works
The assistant receives Hissab's syntax documentation and a calculator tool. It should produce Hissab expressions, call the calculator, and explain the result.
For unfamiliar Hissab syntax, the assistant can fetch focused documentation chunks for categories such as compound units, probability, finance, IP address math, symbolic algebra, and complex numbers before calculating.
For current public facts such as exchange rates, stocks, crypto, weather, or news, the assistant should use realtime search when the selected provider path supports it. If realtime lookup is unavailable or fails, the answer should say so instead of inventing numbers.
Realtime search
The app exposes hosted realtime search through provider adapters that implement it:
- OpenAI — Responses web search tool.
- Anthropic — Messages web search tool.
- Gemini — Google Search grounding.
- Custom endpoints — when the selected SDK / protocol supports the matching hosted search surface.
DeepSeek uses the OpenAI-compatible chat path in the current app, but it is not given the realtime search tool.
Configure external skills
The Hissab app can store custom external skills in Settings. Each skill is
edited as a SKILL.md document and can be enabled or disabled locally in the
browser.
A skill starts with YAML frontmatter, followed by Markdown instructions.
---
name: my-skill
description: One-line description of when to use this skill.
---
Detailed instructions for the assistant go here.
The app requires name and description. The body becomes the skill
instructions.
The parser also accepts allowed-tools or allowed_tools as a comma-separated
list or bracketed list.
---
name: research-helper
description: Use when a question needs source-backed research.
allowed-tools: [realtime_web_search, calculate_with_hissab]
---
Check current facts before calculating with them.
Each saved skill has an On/Off switch. Enabled skills can be included in the assistant context; disabled skills remain saved but are not used.
The app validates frontmatter when you save. Missing frontmatter, missing
name, missing description, or unsupported frontmatter syntax shows an inline
error.
Configure MCP servers
The app can connect to remote MCP servers and make their tools available to the AI assistant. Open Settings, add an MCP server, and fill in:
- Name — the label shown in Settings.
- Server URL — the remote MCP endpoint URL.
- CORS proxy URL — optional prefix added before the server URL.
- Authorization header — optional header value such as
Bearer xxx.
The app uses the streamable HTTP MCP transport. The effective request URL is the optional proxy prefix followed by the server URL.
Use Test connection to connect and list the server's tools. A successful test shows the tool count and names; errors are shown inline on the server card.
Each server has an On/Off switch. Changing the URL, proxy, or authorization header invalidates the existing connection so the next use reconnects with the new settings.
Because the app runs in the browser, remote servers must be reachable from the browser and must satisfy CORS. If a server does not allow browser requests, configure a CORS proxy that you trust.
Attachments
Notebook files can be included with an AI question. Images are passed as image content; text-like files may be passed as text or file/document content depending on the provider adapter.
Attachment support depends on the selected provider and model. If a model does not accept a file type, the provider may reject the request.