Your AI provider.
Your key. Your model.
Bring your own key to Ducalis’s built-in AI chat. Choose a model that fits the task and budget — from organizing feedback to drafting requirements and release notes.
Set up your AI providerSet the provider, API key and default model
Open Integration, Billing → AI → AI provider in your Ducalis organization. This setting powers Ducalis-native AI, including the AI Assistant and messenger bot.
- Provider: choose a supported provider configuration. OpenRouter is one example.
- Base URL (optional): leave the default route in place, or enter a compatible gateway URL if your organization uses one.
- API Key: enter the selected provider’s API key in this settings field.
- Default Model: select an available model suitable for chat and the tools your workflow needs.
- Save: open the built-in AI chat and try a small, clearly scoped task. Check the response and any intended tool action.
This is an organization setting. Check with your team before changing its provider or default model. The settings describe the key as encrypted and shared across the organization; keep it in the API Key field, out of prompts and examples.
New accounts receive AI credits to try the built-in Assistant. Add your own supported provider key to continue when those credits run out. Provider usage charges and limits depend on the configuration you choose.
┌─ BUILT-IN AI CHAT ───────────────┐ │ Read three selected requests │ │ Suggest tags with source links │ ├──────────────────────────────────┤ │ Draft only · review the result │ │ Check facts and tool behavior │ │ Use the model that passes │ └──────────────────────────────────┘
Match the model to the task
Start with a lower-cost model for well-defined work. Move to a more capable model when the task needs deeper reasoning or better writing — and check the evidence in either case.
Routine product work
Suggest tags, categorize requests, sort supplied data or turn selected completed issues into a release-note draft. A lower-cost model can be enough when the input and desired output are clear.
Try a small representative sample first. Check labels, formatting, source links and tool behavior before applying changes to a larger set.
See the release-note workflow →Research and complex drafts
Comparing competitors, resolving conflicting feedback or writing nuanced requirements can justify a more capable reasoning or writing model.
Supply reliable sources, ask it to separate facts from assumptions and review the result. Model price or capability alone does not guarantee factual accuracy.
See reviewed requirements →Spend less on tasks that do not need your strongest model
MiMo-V2.6-Pro, GLM-5.3-Flash and MiMo-V2.6-Flash are concrete value candidates from our 13 October 2026 OpenRouter snapshot below. They combine published independent benchmark results with low token rates. Try them when available through your supported provider configuration.
For a release-note draft, give the chat the selected completed issues, audience and format. Ask for a draft with source links, then check it against the shipped work before publishing.
Try this: “Draft three customer-facing release notes from these completed issues. State what changed and why it matters. Link each note to its issue, mark missing details and do not publish anything.”
Keep a model that consistently passes your checks. If it needs repeated corrections or misses context, change the model and compare the same examples again.
┌─ QUALITY AND TOTAL COST ─────────┐ │ Did the output pass review? │ │ How many retries were needed? │ ├──────────────────────────────────┤ │ Count input and output tokens │ │ Compare the same task samples │ │ Keep it or change the model │ └──────────────────────────────────┘
Strong value candidates, with dated prices and benchmarks
Three options to try for structured product work, alongside flagship API models from Anthropic and OpenAI. These are API token prices, separate from Claude or ChatGPT subscriptions.
// checked 13 October 2026 · OpenRouter catalog rates · USD
| Model | AA Intelligence Index | Input / 1M tokens | Output / 1M tokens | 1,000 input + 200 output |
|---|---|---|---|---|
| MiMo-V2.6-Pro | 46 | $0.435 | $0.87 | 0.061¢ |
| GLM-5.3-Flash | 42 | $0.15 | $0.5 | 0.025¢ |
| MiMo-V2.6-Flash | 38 | $0.14 | $0.28 | 0.020¢ |
| Claude Opus 5.5 | 58 | $4 | $20 | 0.800¢ |
| GPT-6 Astra | 53 | $10 | $50 | 2.000¢ |
The three value candidates cost about 13–41× less than Claude Opus 5.5 and 33–102× less than GPT-6 Astra for this same token budget. That is a price comparison, not equal capability or measured workflow savings.
MiMo-V2.6-Pro has the highest published index of these three value picks; the two Flash models reduce the fixed-budget token cost further. Availability in Ducalis depends on your model list and configuration. Choose based on your own source-grounded task samples and tool reliability.
Compare cost per useful result
The token price is the starting point. Count how much work it takes to get an output your team can use.
Check current token rates
Compare input and output rates for the exact model and provider route. Cached input, reasoning tokens, longer context and additional services can change the bill.
Include retries and review
A cheaper call saves less if it needs several attempts or substantial rewriting. Compare total token usage and the time your team spends correcting the result.
Use the same task samples
Evaluate each model on the same tagging, categorization or drafting examples. Check factual accuracy, formatting and tool reliability before choosing the default.
Token cost estimate: input tokens × input rate + output tokens × output rate, using the provider’s billing units. Add other billed services and all attempts. Ducalis’s default model setting is a manual choice; this page describes a selection process.
Check the rates for your actual provider. Direct-provider prices may differ from an aggregator or gateway route.
Use your chosen model on real feedback
Open the AI Assistant beside your board. Select a customer request, supply the context and ask for requirements, a user story or acceptance criteria.
Review the proposed text with your team. Keep missing information as open questions, confirm the destination and verify any approved change in Ducalis.
┌─ TESTABLE ACCEPTANCE CHECKS ─────┐ │ Given authorized billing user │ │ and a selected date range, │ │ when they export invoices, │ │ then CSV matches that range. │ ├──────────────────────────────────┤ │ Restricted user: no export │ │ No rows: clear empty state │ │ Suggested checks · review │ └──────────────────────────────────┘
BYOK and MCP OAuth are separate
Configure native AI here, or connect the assistant you already use through cloud MCP. Both can coexist.
BYOK for built-in chat
Provider, optional Base URL, API Key and Default Model live under Integration, Billing → AI → AI provider. Your provider supplies model access to Ducalis-native AI.
Open AI provider settings →Cloud MCP through OAuth
Connect Claude, Codex or ChatGPT using the Server URL in personal API, MCP settings and browser OAuth authorization. Your external client supplies its model access. Your BYOK setting remains in place.
Connect your external assistant →The personal REST API token is another credential. It is separate from the provider API key and is not required to complete the MCP OAuth flow.
Choose a first task for the built-in AI chat
Provider, API key and model setup are above. Use these resources to define the task and review the result from your chosen model.
Choose a model. Try one product task.
Start in the built-in chat, review the result and find the balance of quality and cost that works for your team.
Start free