The right model capability for each job: reasoning, code, vision, voice, embeddings, or managed model deployment. Keith can guide the choice, while the router keeps providers healthy.
Outages bring you down
If your single provider has an outage, your whole product stops. No fallback, no warning, no recovery. Just downtime.
Price hikes are mandatory
Provider raises prices? You pay it. No alternatives, no negotiation. Your cost structure is outside your control.
Better models require rewrites
A better model launches from a competitor. Your hardcoded provider means months of migration work to adopt it.
Kaidera abstracts providers behind a single routing layer. Swap, fallback, and route without rebuilding the worker.
Users should not need to memorise model names. Keith can ask what outcome matters, classify the work, and suggest the right capability path. The underlying list is dynamic: admins enable providers and models, and Kaidera routes within that live catalogue.
Architecture choices, product trade-offs, risk analysis, and work that needs deeper judgement.
Implementation, refactoring, code review, tests, and engineering tasks where coding performance matters.
Screenshots, diagrams, invoices, forms, design review, and visual product feedback.
Calls, dictated requests, audio review, transcription, and conversational workflows.
Memory recall, document matching, knowledge retrieval, and similarity search.
Dedicated, geographically controlled, fine-tuned, or self-hosted model programs for unique use cases.
Choose how work is distributed across model sources. Change strategy per worker, per task type, or globally.
Routes to the highest-priority model source in your list. Predictable and consistent — always your first choice unless it fails.
Best for
Regulated environments, model source lock-in preferences, audit trails requiring consistent model use.
Routes to the cheapest model source that meets the task's quality requirements. Cost savings without sacrificing output quality.
Best for
High-volume workloads, documentation generation, bulk processing, cost-sensitive environments.
Routes to the model source with the highest model fit score for this specific task type. Model fit scores updated from real usage data.
Best for
Architecture decisions, complex reasoning, code review, security analysis, critical business logic.
Routes to the model source with the lowest current response time. Continuously tracked. Adapts as model source performance changes.
Best for
Real-time interactions, streaming responses, time-sensitive tasks, user-facing completions.
Define your own routing logic with weighted combinations. Example: 60% priority, 30% price, 10% latency. Route specific task types to specific model sources. Override routing for individual workers.
Watch it live. Simulate a model source failure and see traffic shift automatically. Real-time latency, uptime, and circuit breaker state for every connected source.
OpenAI
GPT-5.2
Latency: 3.2s
Uptime: 99.8%
Anthropic
Claude Opus 4.1
Latency: 2.4s
Uptime: 99.6%
Gemini 2.5 Pro
Latency: 6.8s
Uptime: 97.2%
Fireworks
DeepSeek V3.1
Latency: 1.8s
Uptime: 99.1%
Self-hosted
Qwen3 Coder
Latency: N/A
Last seen: 4m ago
Every request has a fallback chain. Primary → Secondary → Tertiary. If all fail, the circuit breaker opens and the user gets a clean error — not a silent stall.
INCOMING REQUEST
AI worker task requires a model response
ERROR: No model sources available
Clean error returned to user. No silent hang. Circuit breakers remain OPEN until recovery.
Normal operation. Requests flow through. Failures counted but not yet triggering.
Source taken offline after N consecutive failures. Cooldown period begins (60s default).
Single test request allowed. Success = CLOSED again. Failure = OPEN again.
Not all model sources respond at the same speed. With latency-based routing, Kaidera can select the fastest healthy source for the task.
p50 latency in seconds — illustrative benchmark
Kaidera routes to the fastest healthy source when speed matters.For higher-judgement tasks, it can prefer quality or a private deployment instead. The strategy is chosen by task need, not by a static website list.
Prevents cascade failures. When a model source fails repeatedly, the circuit opens automatically. No thundering herd. No resource exhaustion. Clean recovery.
CLOSED
Normal operation
Requests flow through normally. Failures are counted but don't yet block the provider. Default state after recovery.
OPEN
Source halted
After repeated failures, the circuit opens. Requests skip this source immediately while cooldown begins.
HALF-OPEN
Testing recovery
After cooldown, a single test request is allowed through. Success: CLOSED. Failure: OPEN with extended cooldown.
Circuit breakers operate per model source and tenant. One customer's provider issue does not affect other tenants. State transitions are logged and visible in the observability dashboard.
Every provider enforces rate limits. Kaidera tracks them proactively with sliding-window counters — requests per minute (RPM) and tokens per minute (TPM) — so your workers never hit a 429 error.
Sliding window tracks requests per minute per source. When usage gets high, the router shifts new requests to an alternate source before hitting the limit.
Token throughput tracked separately. Large prompts can consume the token budget even when request count is low. Both counters must be green to route.
When a source exceeds 90% of its rate limit window, it receives a routing penalty — effectively deprioritised until capacity frees up. This prevents clustering requests on a nearly-full source.
The model list changes as providers release new capabilities. Kaidera should read from the admin catalogue, then explain choices by capability instead of asking business users to remember model IDs.
Fast models for routine summaries, extraction, classification, and low-risk support work.
Reasoning, coding, vision, voice, embeddings, and document models for specific work types.
Fine-tuned, dedicated, geographically controlled, or self-hosted model programs for specialised use cases.
Access and recommendations are enforced at the routing decision point. If a worker asks for a capability that is not enabled, the router substitutes the best approved option and records the decision for review.
failover time
From failure detection to traffic shift
capability groups
Reasoning, code, vision, voice, embeddings
routing strategies
Priority, Price, Quality, Latency + Custom
Routing picks the right capability. Managed model programs explain what happens when a business needs dedicated, fine-tuned, geographically controlled, or reserved-capacity models.
