Not every task needs the same model. Kaidera gives you granular control over which model each AI worker uses by role, task type, and provider configuration. The catalogue can include frontier reasoning models, efficient models, coding models, vision, voice, embeddings, and managed or isolated deployments as your organisation enables them.
Docs companion
The docs page defines the public dynamic catalog boundary: what model capability data can be shown and what remains admin-only.
Most platforms pick one model for everything. That's wrong in two directions at once.
Using the most expensive frontier model for every task — including updating a comment in a doc file, writing a changelog entry, or running a code review on 5 lines of CSS. You're paying premium prices for zero-value premium reasoning.
Using the fastest efficient model for everything — including architecture decisions, security review, and data model design. You get fast, cheap answers that are subtly wrong in ways that cost you weeks to fix.
The solution isn't one model. It's the right model for each task — automatically.
Assign the best model for each AI worker based on role. Per-worker model assignment is based on what each AI worker is best at. We're constantly evaluating models — including vision models, voice models, embeddings, and open-weight models — and update our recommendations as the landscape evolves. Admins choose what is enabled for the organisation.
Task-type overrides apply regardless of which AI worker is assigned. Architecture tasks use the current frontier reasoning model selected by your admin settings.
Go deeper: assign by task type, not just worker. Architecture decisions can use a frontier reasoning model. Documentation can use an efficient writing model. Code review can use a specialist coding model.
Task-type assignments override per-worker assignments. If you set "architecture → frontier reasoning", every architecture task uses the configured model for that capability, regardless of which AI worker is assigned.
Task types are inferred from the task context, AI worker role, and PREVC phase. No manual tagging required — the routing engine classifies automatically.
Priority order: task-type override → AI worker assignment → org default → platform default. Every task always has a valid model target.
Models are grouped by capability: reasoning, coding, vision, voice, embeddings, search, image generation, and managed model programs. The public website should show examples; the product should read the live catalogue from admin/provider settings.
| Capability | Example models | Best for |
|---|---|---|
Frontier reasoning | GPT-5.2Claude Opus 4.1Gemini 2.5 Pro | Strategy, architecture, complex review |
Build and code | Claude Sonnet 4DeepSeek V3.1Qwen3 Coder | Implementation, refactor, review |
Efficient work | GPT-5 miniGemini FlashQwen3 instruct | Summaries, docs, routine extraction |
Vision and documents | GPT visionClaude visionGemini vision | Screenshots, forms, diagrams, visual QA |
Voice and realtime | Realtime audioTranscriptionSpeech generation | Calls, dictated requests, voice support |
Embeddings and retrieval | text-embedding-3-largeprovider embeddingsprivate embeddings | Search, memory recall, document matching |
Private and fine-tuned | custom fine-tunededicated endpointself-hosted Qwen class | Regulated or specialised workflows |
These examples are illustrative. The production catalogue should come from admin/provider configuration so users see the models currently enabled for their organisation.
Organisations can connect approved provider accounts, enable model families, and decide which capabilities appear to AI workers. Kaidera provides recommendations, but the configured catalogue remains the source of truth.
Route routine tasks to efficient models. Reserve premium models for what they're worth.
Organisations can connect approved provider API keys. Use your negotiated rates, your own rate limits, and your existing provider relationships — while keeping Kaidera's routing, monitoring, and model recommendations.
Store your OpenAI, Anthropic, or Google API keys. Encrypted in cloud secret manager — never in the database.
When you add a key, Kaidera probes the provider to discover which models you have access to. No manual config needed.
BYO key requests bypass Kaidera's model markup. You pay your provider directly at your negotiated rate.
External subscriptions — Connect your existing approved provider accounts. Kaidera detects available model capabilities and routes accordingly.
Model assignment is the what. Model routing is the how — capability matching, provider health, fallback chains, and live recommendations.
