specification: API Commons ChangeLog
specificationVersion: '0.1'
provider: LaunchDarkly
providerId: launchdarkly
generated: '2026-08-27'
method: searched
source: https://launchdarkly.com/changelog/
checked: '2026-08-27'
http_status: 200
docs: https://launchdarkly.com/changelog/
note: >-
  LaunchDarkly maintains TWO changelogs and they answer different questions. The product
  changelog below is dated, attributed and filterable by topic. The API VERSION changelog
  — the one that tells an integrator whether their code will break — lives inside
  info.description of https://app.launchdarkly.com/api/v2/openapi.json and is recorded in
  lifecycle/launchdarkly-lifecycle.yml with its EOL dates.
scheme:
  product: dated entries, attributed to a named author, tagged by topic
  api: date-versioned (LD-API-Version, yyyymmdd), with a per-version EOL date
  current_api_version: '20240415'
surfaces:
  - name: Product Updates
    url: https://launchdarkly.com/changelog/
    dated: true
    attributed: true
    topics: [AI Engineering, Feature Flags, Experimentation, Guarded Release, Product Analytics]
    archive_depth: 'Monthly filters back to May 2025 (16 months) as of 2026-08-27'
    machine_readable: false
    note: >-
      Server-rendered for only the first page; older entries load client-side, so only
      the most recent batch is harvestable without a browser. No RSS or JSON feed for
      the changelog was found.
  - name: API version changelog
    url: https://app.launchdarkly.com/api/v2/openapi.json
    location: info.description
    dated: true
    machine_readable: true
    breaking_changes_flagged: true
    note: The only place breaking API changes and EOL dates are published together.
  - name: Blog
    url: https://launchdarkly.com/blog/feed/
    http_status: 200
    content_type: application/xml
entries:
  - date: '2026-08-26'
    topic: AI Engineering
    author: Paul Loeb
    title: Use AgentControl from your Go and Java applications
    kind: addition
    breaking: false
    highlights: >-
      AI SDKs released for Go and Java. Prompts, model settings and targeting rules move
      out of code into AgentControl configs; both SDKs report generation counts, token
      usage, latency and errors, and both run multi-step agent graphs.
  - date: '2026-08-26'
    topic: AI Engineering
    author: Paul Loeb
    title: Serve different models to different users through your LiteLLM gateway
    kind: addition
    breaking: false
    highlights: >-
      Per-request model, prompt and parameter selection through an existing LiteLLM
      gateway, without editing YAML, restarting the proxy or deploying code. Changes are
      attributed and approval-gated like any flag change.
  - date: '2026-08-26'
    topic: AI Engineering
    author: Paul Loeb
    title: AgentControl now supports EU data residency
    kind: addition
    breaking: false
    highlights: >-
      Prompts, configs, variations and targeting rules can now be managed in the EU
      instance, with the same approval workflows and audit history as feature flags.
      Model choices are limited to providers approved for the region.
  - date: '2026-08-26'
    topic: AI Engineering
    author: Paul Loeb
    title: Read a full agent run as a single conversation
    kind: addition
    breaking: false
  - date: '2026-08-26'
    topic: AI Engineering
    author: Paul Loeb
    title: Use your Amazon Bedrock models in AgentControl
    kind: addition
    breaking: false
window:
  from: '2026-08-26'
  to: '2026-08-26'
  entries_captured: 5
  note: >-
    Only the server-rendered first page was harvestable; the site paginates the rest with
    JavaScript. All five most recent entries are AgentControl/AI, which is itself the
    finding — LaunchDarkly's shipping velocity in August 2026 is almost entirely on the
    agent-control surface, not on feature flags.
recent_api_changes:
  last_breaking_version: '20240415'
  last_breaking_date: '2024-04-15'
  note: >-
    No breaking REST API version has shipped in roughly 28 months. The product surface
    moves weekly; the API contract's version axis has been still since April 2024, and
    new capability arrives as new (beta) tags inside the current version rather than as a
    new version.