Cheap mode
Background reasoning
Use lower-cost requests for summaries, notes, or internal agent updates.
Build memory-aware AI agents with one API for execution, routing, and provider fallback.
A good agent stack usually has two parts: a memory system that decides what to remember, and an execution layer that turns that context into reliable model calls. BrainAPI helps with the second part so your agent code stays clean while your memory architecture evolves.
Why this matters for agents
Agents are more fragile than chatbots because they often need to recall previous steps, keep tool results, summarize work, and route requests differently based on cost or latency. If every provider requires a new payload shape, your agent orchestration becomes the bottleneck.
- Store the agent's memory state in your own durable system.
- Fetch only the context the next step needs.
- Send the final execution request through BrainAPI.
BrainAPI's role in an agent stack
Fast mode
Interactive steps
Respond quickly when the user is waiting for the next action or update.
Best mode
High-value completions
Route premium prompts when quality matters more than latency or cost.
Reference architecture
- Agent receives a task and loads the current session state.
- Your app retrieves relevant memory from its own store.
- Your orchestrator composes the step prompt or tool result.
- BrainAPI handles the AI request, fallback, and normalized response.
- Your app saves a new summary or memory snapshot for later steps.
Example request
Unified agent requestcURL
curl -X POST https://api.brainapi.site/api/v1/ai \
-H "X-API-Key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"type": "text",
"input": "Use the attached agent summary and produce the next action.",
"mode": "auto",
"max_output_tokens": 240
}'
Related guides
Need a concrete implementation path?
Read AI Memory API for Chatbots for architecture choices or OpenAI Memory API Alternative for product comparison positioning.