Skip to content

Introducing Cortex

Give the agents you already use the knowledge they need before they act.

Learn more

Modules hold rules. Issues name work. SemDAGs map dependencies. A flash reads needed context.

A 35-Module SemDAG for a fictional agent evaluation and deployment platform. Tab enters the graph once. Use arrow keys to move between nodes, Enter or Space to activate a node, and Tab to leave.When options conflict, rank safety first, verified reliability second, then the simplest reversible design; never trade a higher-ranked property for speed or novelty.valuesOne person keeps one identity across verified sign-in methods; ownership, approvals, and audit authorship remain attached after credentials change.idEvery membership, project, agent version, deployment, and subscription belongs to one account; authority and data never cross accounts.accountEach token selects one account-local actor, audience, and expiry; account switching mints a distinct token.account-sessionEvery account-owned row uses composite account keys and foreign keys; each transaction runs under an account-bound database policy.dbA project is one account-owned boundary for agent versions, evaluation suites, and deployment aliases.projectAn agent version immutably pins its model, instructions, tool schemas, retrieval snapshot, and output contract; any change creates a new version.agent-versionEach command derives actor and account from its session, then checks the action against account or project roles; resource identifiers grant no authority.authorizationAn account publishes immutable evaluation-suite versions that define fixtures, allowed tools, invariants, scorers, thresholds, and blocking conditions.evaluation-suiteAn evaluation pins one agent version and one evaluation-suite version; changing either creates a new evaluation.evaluation-contractEach gateway revision fixes provider routing, model mappings, retryable failures, timeouts, cost limits, and normalized response and usage schemas.model-gatewayEvery evidence item names one completed run attempt, input digest, observed tool call or output, collection method, and collector version.run-evidenceA completed run publishes one immutable result for its evaluation contract, with scorer outputs, evidence, threshold decisions, and a stable deduplication key.evaluation-resultOne evaluation contract and gateway revision own one durable run; attempts checkpoint inputs, tool calls, costs, outputs, and terminal state.run-ledgerOnly authorized human approvals count; each names one candidate and immutable evaluation result, and any superseding result or failed required threshold expires it.promotion-approvalRollout opens only when the candidate equals the approved version, that approval remains valid, and its required results have no unresolved blockers.deployment-gateEach project environment resolves one gate-approved agent version; alias history is immutable, and superseded assignments never reactivate.deployment-aliasThe evaluation runtime resumes one durable run and publishes its evidence and result atomically after all fixtures and tool effects settle.evaluation-runtimeShared tokens fix hierarchy, spacing, contrast, motion, focus, and 44-pixel targets across every public and product surface.design-systemThe studio keeps version changes, evaluation evidence, run state, approval, deployment gate, and proposed rollout together.agent-studioSettings shows active and draft suites and deployment aliases; suite changes affect only evaluations created afterward.project-settingsA billable evaluation is one contract's first immutable result; operational health canaries are excluded, and retries, resumed attempts, or gateway changes never add charges.billing-policyEach immutable plan version fixes price, evaluation allowance, concurrency, result retention, and project limit; published versions never change.plan-catalogUsage records one debit for each billable evaluation and reverses it only through an explicit correction event.usage-ledgerPayment-provider events map to one account and apply once by authoritative version; effective state retains its plan version until an explicit change.subscription-ledgerEntitlements combine project role, subscription state, and usage: grace preserves reads, exhausted allowance blocks new runs, and payment grants no project authority.entitlementsEvery material action appends account, actor, action, target, rule version, outcome, and time; export requires audit permission and excludes prompts, outputs, and secrets.audit-logTeams compare candidate agent behavior on locked tasks before deciding which exact version may reach production.positioningMessages show the pinned candidate, named evaluation, human approval, and version-matched deployment gate; changing the candidate closes that gate.marketingThe page demonstrates one candidate from locked task through cited evidence, human approval, and exact-version deployment result.landing-pageThe page renders commercial limits and prices verbatim, explains the billable-evaluation unit, and never authors entitlement rules locally.pricing-pageEvaluation-suite guides pin suite, agent, gateway, and configuration versions; every snippet runs in CI against a disposable project before publication.evaluation-suite-docsA nonbillable test-project canary requests one forbidden tool call through its alias; require blocked action, zero effects, complete trace, and result within 60 seconds.agent-path-healthOne manifest pins compatible schema, gateway, runtime, and health checks; old runs finish pinned, a canary proves the path, and rollback preserves runs.release-safetySuspend aliases, revoke exposed credentials, bound account and run IDs, preserve traces, notify owners, and roll back only when evidence implicates the current release.incident-response

WorkspaceOne workspace for your team and agents

An isolated home for your organization, its Socra products, and every person and agent working across them.

Learn more
Workspace

One organization

Every Socra product shares the same people, agents, and account controls.

  • Cortex
  • Journeys
  • Admin
  • Your team
  • Your agents
  • AdminOne place to manage
  • Organization controlsMembers, groups, and roles
  • Shared billingOne account across products

Built for developers and agents

Developers set the standard. Agents receive it before they act.

JourneysAugment Your Thinking with AI Agents

For content, vision, planning, and outside-code problems, Socra owns the reasoning and introduces one specialist only when useful.

Learn more
Go to Journeys

  • @jobs

    Visionary

  • @davinci

    Inventor

  • @socrates

    Philosopher

  • @turing

    Developer

  • @einstein

    Genius

  • @tesla

    Engineer

  • @curie

    Researcher

  • @holmes

    Detective

  • @feynman

    Teacher

CloudBuild, run, connect, and secure software through small services that work together

Source becomes an immutable artifact. A runtime serves it through a stable HTTPS route.

Learn more
Go to Cloud Console
socra run deploy
BuildSource to image
RegistryImmutable artifact
RunPreview to production
DomainsStable HTTPS
Recorded project state
August 30, 2026
Project
Name hidden
Container Registry
Enabled
Sandbox
Enabled
  • SandboxAgent environments
  • DatabaseManaged databases
  • SecretsWrite-only values
  • AnalyticsEvents and reports

Connect Cortex to your coding agent

Start with one Module. Its dependency tree gives the agent the rules and reasons behind your code.

Try Cortex