Chat memory versus live evidence
Run the same current-events request with and without tools.
Start lessonLearn how models search, create media, understand documents, work with files, coordinate agents, and finish repeatable tasks. Change one control at a time and see what actually happens.
Your guidance
contextual
Use question-mark controls whenever you want an explanation.
Change learning settingsAsk the same current-events question twice. Simple mode can only use the model's existing knowledge. Agent mode can search the web and show its sources. Compare freshness, evidence, time, and cost.
Review every source before relying on the result. Tool access does not guarantee that every claim is correct.
Guidance preference
This changes explanations, not permissions. You keep access to the same supported controls.
Practical lessons
Each lesson connects explanation to a real MaiaChat control, experiment, or evidence surface.
Run the same current-events request with and without tools.
Start lessonUse MaiaChat’s Comparison Lab to isolate the model or instruction profile.
Start lessonConnect a provider while protecting credentials and controlling spend.
Start lessonLearn what belongs in temporary context, document retrieval, or durable memory.
Start lessonTurn a general model into a bounded specialist and inspect what changed.
Start lessonUse iterative agents without creating an endless or expensive process.
Start lessonMove from a hosted exercise to project work with approvals, diffs, and rollback.
Start lessonLearn when specialist collaboration improves a result—and when it only adds cost.
Start lessonTurn a successful workflow into scheduled work with clear limits and notifications.
Start lessonCapability path
Follow the path or enter anywhere. Nothing is locked, and every lesson should end with something observable.
Understand what an LLM is and why the model you choose changes the result.
Outcome: Compare two models with confidence.
Move from temporary conversation context to grounded documents and durable knowledge.
Outcome: Know what MaiaChat can recall and why.
See the difference between a model answering from memory and an agent using evidence.
Outcome: Run a grounded web-search experiment.
Explore models made for visual creation, image understanding, transcription, and speech.
Outcome: Choose a model for a non-text task.
Progress from a safe hosted workspace to inspected, approved, and verified project changes.
Outcome: Complete a reversible project change.
Learn how roles, system prompts, models, tools, and parameters shape reliable specialists.
Outcome: Build and compare a specialist agent.
Replace 'the model says it is done' with observable checks and explicit stopping conditions.
Outcome: Run a bounded task with evidence.
Compare one agent with sequential, parallel, hierarchical, consensus, and MOA collaboration.
Outcome: Explain the quality, latency, and cost trade-off.
Turn successful work into reusable skills, schedules, workflows, and notifications.
Outcome: Run a safe repeatable routine.
An API key lets MaiaChat use models billed to your provider account. Treat it like a password: create it on the provider's official site, store it only in the Settings key field, monitor usage, and revoke it if exposed.
Never paste a key into a chat message, lesson, screenshot, source file, or public repository. MaiaChat's Settings page masks stored keys and validates them without returning the full secret.
Learn which models create images and which understand them.
Compare transcription, speech generation, and voice conversation.
See how instructions, tools, loops, and orchestrators change work.