OSS-first docs

These docs teach the open system first: contracts, generated surfaces, runtimes, governance, and incremental adoption. Studio shows up as the operating layer on top, not as the source of truth.

AI index

Agent ecosystem · Self-improvement

Let agents improve through evaluated, reversible promotion.

Self-improvement is a governed release process for agent behavior—not permission for an agent to rewrite its own policy, authority, or production boundary.

Improvement lifecycle

feedback → propose → evaluate → approve → canary
         → apply → observe → promote
                           ↘ rollback

Create the governed runtime

import { GovernedAiImprovementRuntime } from
  "@lssm-tech/lib.ai-agent/improvement-loop";
import { PostgresAiImprovementWorkOrderStore } from
  "@lssm-tech/integration.agent-durable-store/postgres";

const improvement = new GovernedAiImprovementRuntime({
  store: new PostgresAiImprovementWorkOrderStore({ database, tenantId }),
  policyResolver,
  adapter: { propose, evaluate, canary, apply, rollback },
  clock,
  jobQueue,
});

const work = await improvement.enqueueWork({
  scope: { tenantId, workspaceId, agentId },
  dedupeKey: feedbackBatchDigest,
  expiresAt,
  feedback: consentedRedactedFeedback,
});

Required controls

Proposal-only default

New deployments create evidence-bound proposals and stop before mutation.

Digest-bound approval

Any proposal, evidence, policy, or patch change invalidates the approval.

Canary and observation

Promotion waits for bounded traffic and explicit quality, safety, cost, and latency evidence.

Idempotent effects

Apply and rollback use stable effect identifiers and durable receipts.

Policy re-resolution

Policy is resolved again before every stage, including delayed jobs.

Kill switches and expiry

Operators can halt queued or active work; stale work orders fail closed.

Feedback is governed data

Collect only consented, tenant-scoped, redacted feedback. Never let raw prompts, credentials, private repositories, or unsupported telemetry become an unreviewed training or mutation source.

AI agent package reference

Reference application guide

OSS docsagentsStart with OSS. Adopt Studio when you want the operating layer.
Previous

Multi-agent collaboration

Coordinate subagents, elastic teams, multiplayer sessions, leases, claims, mailboxes, handoffs, replay, and human intervention.

Next

Durable agent runtime

Run durable memory, encrypted storage, continuous loops, schedulers, workers, retries, cancellation, and dead-letter handling.

Why ContractSpec

Keep educational and comparison content reachable without letting it define the primary OSS learning path.

ManifestoContract-first APISpec-driven development