Skip to main content

C-13 Resource Limits

Brick C: Agents, autonomy, execution. French version: C-13-resource-limits

1. Role and responsibilities​

Bounds consumption per run, agent and tenant: Kubernetes quotas (CPU, memory) and runtime limits (steps, tokens, cost, duration, concurrency).

  • Reserve a quota before the run (brick H quotas)
  • Apply ResourceQuota / LimitRange
  • Limit concurrent runs per agent

2. Position in the architecture and interactions​

  • Callers: Agent Runtime, Sandbox, FinOps.
  • Execution path: Synchronous at start, continuous check in the loop.
  • Every interaction goes through the Control Plane API v1 (/v1) behind the API Gateway (D-14).

3. Open source component and developed part​

  • Open source: Kubernetes ResourceQuota, LimitRange
  • Developed: Runtime limits and quota reservation.
  • External integration through an adapter eai/integrations/<system>.py with an in-process fallback when the URL is not configured.

4. Data model​

Tables (schema eai, tenant_id RLS):

  • eai.runs (budgets)
  • eai.quotas

Common columns: id uuid, tenant_id uuid NOT NULL, created_at, updated_at; RLS policy current_setting('app.tenant_id', true)::uuid.

5. Contract​

  • Applied contract: agent-manifest + execution-contract
  • JSON Schema: contracts/v1/agent-manifest.schema.json, contracts/v1/execution-contract.schema.json
  • Key fields: spec.budgets, budgets.max_cost, budgets.max_duration_s
  • Validation: POST /v1/contracts/{name}/validate; manifest apply: POST /v1/manifests/apply.

6. API endpoints​

  • POST /v1/quotas/reserve
  • POST /v1/agents/{id}/runs
  • PATCH /v1/agents/{id}

Request PATCH /v1/agents/{id}:

{
  "budgets": {
    "max_steps": 30,
    "max_concurrent_runs": 5,
    "max_cost": {
      "amount": "2.00",
      "currency": "EUR"
    }
  }
}

Response:

{
  "id": "agt_support",
  "version": 4,
  "budgets": {
    "max_steps": 30,
    "max_concurrent_runs": 5
  }
}

7. SDK examples​

Python:

from enterprise_ai import EnterpriseAI

client = EnterpriseAI(base_url="<control-plane-url>", api_key="eai_xxx_secret")
client.agents.run("agt_support", "Triage the open tickets", budgets={"max_tool_calls": 30})

TypeScript:

import { EnterpriseAI } from "@enterprise-ai/sdk";

const client = new EnterpriseAI({ baseUrl: "<control-plane-url>", apiKey: process.env.EAI_API_KEY });
await client.agents.run("agt_support", { task: { objective: "Triage INC-42" }, budgets: { max_steps: 30 } });

Java:

EnterpriseAIClient client = EnterpriseAIClient.builder()
    .baseUrl("<control-plane-url>")
    .apiKey(System.getenv("EAI_API_KEY"))
    .build();
client.runs().start("agt_support", Map.of("task", "Summarize INC-42", "budgets", Map.of("max_steps", 30)));

Go:

c, _ := eai.NewClient(eai.WithBaseURL("<control-plane-url>"), eai.WithAPIKey(os.Getenv("EAI_API_KEY")))
c.Runs.Start(ctx, "agt_support", map[string]any{"task": "Summarize INC-42", "budgets": map[string]any{"max_steps": 30}})

8. Configuration​

  • Environment variables: EAI_DATABASE_URL
  • Helm values keys: controlPlane.resources, sandbox.runtimeClasses

9. Security controls​

  • Limits cannot be changed by the agent itself
  • Every request carries the tenant resolved from the credential; PostgreSQL RLS
  • Every mutation and policy decision writes a hash-chained audit event

10. Observability​

  • Prometheus metrics: eai_budget_exceeded_total{kind}, eai_runs_concurrent
  • Traces: OpenTelemetry spans with GenAI semantic conventions (gen_ai.operation.name, gen_ai.agent.id, gen_ai.tool.name), attributes eai.tenant_id, eai.run_id, traceparent propagation.
  • Audit actions: run.transition

11. FinOps units​

  • Quota reservation then commit / release.
  • Emitted as Usage Event (POST /v1/finops/usage) with tenant, project, agent and run.

12. Failure modes and resilience​

  • Quota unavailable -> 429 quota_exceeded
  • Automatic release at run end
  • Normalized errors (error envelope, message_i18n en/fr)

13. Standalone adoption (client of a client)​

  • enterprise_ai.agent + enterprise_ai.quotas.
  • SDK: enterprise_ai.agent (AgentsClient, RunsClient, SessionsClient)
  • Minimal deployment: Helm chart with controlPlane.enabled=true and PostgreSQL; disabled external components use their in-process fallback.

14. Acceptance tests​

  1. 6th concurrent run with limit 5 -> 429
  2. Cross-tenant access refused
  3. Error returned in fr with Accept-Language: fr