Context before the first line
Before an agent edits a service, Conera hands it live production memory: dependencies, latency profile, and prior incidents so the plan is grounded before code is written.
> user: add a retry policy to the payment client
⏺ Conera · service context (payment-service)
⏺ Conera · similar changes & incidents
agent: "Before writing this: payment-service is latency-sensitive at peak, and an aggressive retry here caused inc_31 in May by amplifying load during a Redis failover. I'll use exponential backoff with a cap, matching the recorded fix from that incident."
> user: add a sync batch job to invoice-worker
⏺ Conera · runtime constraints (invoice-worker · Azure Functions)
⏺ Conera · related incidents
agent: "This function's plan enforces a 230s timeout. A previous long-running batch here was killed mid-write (inc_44). I'll split into smaller batches with checkpointing instead of one sync job."
Runtime constraints in the session
Cold starts, plan timeouts, and platform limits are injected while the agent works—so it doesn't propose something your cloud will kill at runtime.
Past breakage shapes the diff
When similar changes failed before, Conera surfaces that history so the agent copies the fix that worked—not the pattern that caused the outage.
> user: increase Redis connection pool in checkout-service
⏺ Conera · similar changes & incidents
⏺ Conera · recorded fix (inc_18)
agent: "inc_18: pool raised 20→80 without client timeouts; Redis CPU spiked and checkout p99 hit 4.1s. Recorded fix was pool 32, idle timeout 30s, circuit breaker on Redis errors. I'll apply that config—not a blind bump."
> user: call inventory-service from checkout to check stock
⏺ Conera · service topology (checkout-service → inventory-service)
⏺ Conera · runtime & dependencies
agent: "checkout runs on AKS; inventory exposes gRPC over mTLS on the mesh—not REST. inventory depends on PostgreSQL via Azure Private Endpoint. I'll add a gRPC client with the mesh identity and the recorded deadline—no HTTP client, no public DNS."
System details you don't have to brief
Protocol, cloud runtime, and what the service depends on—Conera already knows where the change will run, so the agent implements against reality without you pasting architecture into the prompt.