Changelog
What we shipped, when we shipped it.
v0.19.1
private beta- Fixed Webhook retry attempt 4 was firing at 15min instead of 30min. Now matches the docs.
- Fixed duration_ms was null on sub-millisecond trace steps. Returns 0 now instead of null.
- Improved X-RateLimit-Reset header added -- Unix timestamp so you know exactly when your quota resets
v0.19.0
private beta- Added Parallel workflow steps -- independent steps run concurrently instead of waiting in line
- Added task.progress webhook event. Fires during long-running tasks with a progress_pct field.
- Improved Token refresh kicks in 60s before expiry now, not after it fails
- Removed /v1/tasks/dry-run endpoint removed. Use mode: "test" on task creation instead (available since v0.16.0).
v0.18.0
private beta- Improved Execution memory splits per-user patterns from global ones. Your tasks learn from your usage, not everyone's.
- Fixed OAuth incremental scope requests weren't firing for custom tools
v0.17.2
private beta- Fixed Webhook signature verification broke on payloads with unicode. Was hashing the escaped bytes, not the raw string.
- Fixed GET /v1/tasks ignored offset when you also passed a status filter
- Improved workflow_depth_exceeded errors now tell you the step count and which tool went too deep
v0.17.1
private beta- Fixed Idempotency-Key header matching was case-sensitive. Now case-insensitive per HTTP spec.
- Fixed Test mode returned 500 on tasks with no matching tools instead of a clear validation error
v0.16.0
private beta- Added Idempotency keys -- pass an Idempotency-Key header on task creation for safe retries
- Added Test mode (mode: "test"). Validates your prompt and shows which tools would be selected, without actually running anything.
- Improved Workflow generation ~35% faster on multi-tool tasks. Tool selection happens in parallel now.
- Removed /v1/tasks/dry-run endpoint deprecated. Use mode: "test" instead.
v0.15.0
private beta- Added Custom tools -- POST /v1/tools to register your own APIs and internal services
- Added Auth modes for custom tools: API key, bearer token, OAuth 2.0, or none
- Improved Tool selection accuracy up ~18%. Better matching between what you ask for and what tools can do.
v0.14.0
private beta- Improved Schema drift detection picks up field renames and type changes, not just missing fields
- Improved Task traces show inline on the dashboard now, right next to the execution timeline
- Fixed Trained task memory was getting wiped on region migration. Missed a replication flag.
- Fixed Race condition: two concurrent tasks could both pop the OAuth consent screen for the same service
v0.13.0
private beta- Added Trained Tasks. Every run teaches the system something. Over time your tasks get faster and break less.
- Added Built-in tools ship pre-trained so they don't start from scratch
- Improved Token expiry recovery: ~200ms, down from 1.2s
v0.12.1
private beta- Fixed Webhooks hung forever if your endpoint returned a 301
- Fixed Cancelling a task didn't stop tool calls that were already in flight
- Improved 429 retry backoff actually reads the Retry-After header from third-party APIs now
v0.12.0
private beta- Added Task traces -- GET /v1/tasks/:id/trace for the full step-by-step execution with timing data
- Added Trace steps now include notes when recovery or adaptation happened, so you can see why a step was retried or skipped
- Improved Dashboard redesign. Real-time task status, filterable history, less clutter.
v0.11.0
private beta- Added Per-task timeout -- set timeout_ms on task creation to cap how long a task can run
- Improved Auth context is shared across workflow steps now. No more re-authenticating between each tool call.
- Improved Error responses include a request_id field for faster support debugging
v0.10.0
private beta- Added Self-generating workflows. No more predefined DAGs -- Ploton figures out the execution plan at runtime.
- Added Mid-execution adaptation -- steps get added, skipped, or retried based on what actually happens
- Improved Token expiry, rate limits, schema changes -- all handled in-flight now without failing the task
- Removed callback_url field on task creation deprecated. Register webhooks via the dashboard instead.
v0.9.0
private beta- Improved Ambiguous prompts now resolve to the right service based on what you have connected
- Improved POST /v1/tasks response includes the selected tool name so you know what ran
- Fixed Webhook timestamps were in server local time, not UTC
- Fixed Metadata keys with null values were silently dropped. They're preserved now.
v0.8.0
private beta- Added Webhook signature verification (HMAC-SHA256, X-Ploton-Signature header)
- Added Webhook retries -- exponential backoff, 5 attempts over 24h
- Added task.waiting event fires when OAuth consent is needed. Includes auth_url so you can surface the prompt.
- Improved Webhook payloads include a timestamp field now (ISO 8601)
v0.7.0
private beta- Added metadata field on tasks. Pass whatever you want -- order IDs, user IDs, correlation keys.
- Improved OAuth consent supports incremental scopes so users don't get a wall of permissions upfront
v0.6.1
private beta- Fixed Resolved an access control issue where task list pagination could return records from other accounts under specific offset values. Patched within 2 hours of discovery. No evidence of exploitation.
- Fixed OAuth refresh tokens now encrypted at rest (AES-256). Previously stored in plaintext.
- Improved API errors include a suggestion field with what to try next
v0.6.0
private beta- Added Task API -- create, get, list, cancel via REST
- Added Webhooks for task.complete and task.failed
- Added Managed OAuth -- we handle token refresh so your agents don't have to
- Added Test keys (pk_test_) and live keys (pk_live_). Same API, same behavior, different environments.
v0.5.0
private beta- Added Private beta. API live at api.ploton.ai/v1 for early access partners.
- Added Basic webhook delivery on task completion
- Added Dashboard with API key management and webhook config
- Improved Beta rate limit: 100 req/min