Aegis
Pulse never fails.
Aegis is the resilience layer that ensures Pulse always delivers a result. Fifteen interlocking defense mechanisms — from tool fallback chains to model failover to circuit breakers — guarantee that no matter what goes wrong, your request is fulfilled.
The fundamental principle of Aegis is absolute: Pulse never says 'I can't'. Between dedicated tools, alternative APIs, web search, browser automation, model failover, circuit breakers, progressive retries, and a knowledge-based last resort — there is nothing Pulse cannot accomplish.
Aegis operates at three levels: tool-level resilience (fallback chains, circuit breakers), loop-level guards (empty response retry, refusal detection, relevance checking), and system-level protection (model failover, pipeline retry, rate limit absorption). The user never sees an error — only a result.
Resilience Layers
Tool Fallback Chain
Five cascading layers per tool: dedicated API, alternative provider, web search, web fetch, and browser automation. Each layer activates automatically.
Circuit Breaker
Tools that fail consecutively are temporarily disabled. After a cooldown, a half-open probe tests recovery before restoring full traffic.
Model Failover
If the primary AI provider is down, Aegis automatically switches to a fallback model. Conversations continue without interruption.
Empty Response Guard
If the AI returns an empty or blank response, Aegis retries with progressively urgent prompts — up to three attempts before escalating.
Refusal Guard
Detects phrases like 'I can't' or 'no puedo' in five languages. Triggers an automatic retry with the Absolute Rule: always find a way.
Relevance Guard
Detects off-topic or 'found nothing' responses and re-prompts the AI to stay focused on the original request.
Pipeline Retry
The agent pipeline retries up to three times, simplifying the configuration on each attempt to maximize the chance of success.
Knowledge Last Resort
If all tools and retries fail, Aegis makes a final LLM call with no tools — relying purely on the model's built-in knowledge to help.
Rate Limit Absorption
When a provider returns a rate limit error, Aegis auto-waits up to 5 seconds and retries — the user never sees the delay as an error.
Guaranteed Resolution
Pulse never says 'I can't'. Forbidden phrases are enforced at the system prompt level in five languages. Every request gets a real solution.
Proactive Token Refresh
OAuth tokens are refreshed 5 minutes before expiry — not after failure. Central 401 detection auto-invalidates stale caches across all tools.
Channel Health Monitor
Every 60 seconds, Aegis checks all connected channels. Dropped connections are auto-reconnected with exponential backoff and circuit breaker.
15 Layers, Zero Failures
Tool fallback chains, circuit breakers, model failover, pipeline retries, empty response detection, tool announcement detection, refusal guard, relevance guard, sub-agent error normalization, knowledge last resort, rate limit absorption, context compression with extension, proactive OAuth token refresh, channel auto-reconnection, and the Absolute Rule — all working together so you never see an error.
How Aegis protects every request
Tool executed
The primary tool runs with proper authentication. If it succeeds, the result is returned. If not, Aegis activates.
Fallback chain engaged
Alternative APIs, web search, web fetch, and browser automation are tried in sequence. The circuit breaker skips known-failing tools.
Loop guards monitor
Empty response, refusal, tool announcement, and relevance guards check the AI's output. Failed checks trigger automatic re-prompts.
Model failover if needed
If the primary provider is down, Aegis switches to the fallback model. The pipeline retries up to three times with simplified config.
Knowledge last resort
As the final safety net, a no-tools LLM call uses the model's built-in knowledge. Pulse always delivers a result.