Pilot Changelog

All notable public changes to Pilot are documented here. Entries focus on self-hosting, APIs, governance, founder workflows, connectors, security, and documentation impact.

[1.3.0] - 2026-04-24

Added

• Long-running autonomous execution checkpoint support for task runs, including checkpoint state and stalled-run detection.
• Cost attribution dashboard material for spend, cache savings, monthly burn projection, cache hit rate, and per-workspace or per-operator views.
• Skills marketplace client for installing verified skill archives into a local Pilot skill directory.
• Ollama provider support for local/self-hosted inference when no cloud provider key is configured.

Changed

• Pilot version updated to 1.3.0.
• LLM configuration supports self-hosted Ollama endpoint and model settings.
• Environment examples include self-hosted inference and skills registry configuration.

[1.2.1] - 2026-04-24

Fixed

• A2A task dispatch now routes into the governed orchestrator instead of returning canned responses.
• Evidence-pack writes run L1 conformance validation.
• Tool output sanitization now runs before untrusted tool results return to the model context.

Added

• Subagent certification CLI for validating evidence packs.
• A2A gateway tests, connector unit tests, and sanitizer tests.
• Operator-ready environment examples for A2A and optional PDF parsing.

[1.2.0] - 2026-04-20

Added

• Connector breadth across Slack, Notion, Linear, Stripe read-only, Google Calendar, and HubSpot.
• Vision and PDF ingestion helpers with parse_pdf and analyze_image tools.
• Agent2Agent discovery and JSON-RPC endpoint with bearer authentication.
• L1/L2 conformance validators for receipt and evidence-pack checking.

Changed

• Built-in orchestrator tools expanded from 27 to 47.
• Default registered connectors expanded from 8 to 13.

[1.1.0] - 2026-04-19

Added

• MCP consumer and provider support, including a Pilot MCP server for DB-backed tools.
• Compliance framework overlays, attestations, and governance UI hooks.
• Sandbox abstraction for generated-code execution checks.
• Optional Langfuse and Braintrust observability hooks.
• Skill loader for SKILL.md based operator behavior.
• HELM endpoint integration, budget/cost surfaces, and proof graph support.
• Threat scanning and sanitizer tests for untrusted scraping output.
• Anthropic prompt-caching telemetry.
• Live conduct streaming through server-sent events.

[1.0.0] - 2026-04-19

Added

• Connector token refresh infrastructure.
• CI/CD hardening with SBOM, provenance, vulnerability scanning, secret scanning, and restore drill workflow.
• Observability alerts, dashboards, Sentry/OpenTelemetry documentation, and cost/governance views.
• DigitalOcean production deployment path with Pilot, HELM sidecar, PostgreSQL, Caddy, and backup scheduling.
• Web dashboard mode pages, governance DAG, Telegram command routing, E2E tests, and load-test baseline.

[0.1.0] - 2026-04-16

Added

• HELM governance sidecar integration.
• Multi-tenant workspace isolation with encrypted secret storage.
• Scoring, semantic deduplication, clustering, decision court, build, launch, and apply workflows.
• Magic-link and Telegram authentication.
• DigitalOcean deployment configuration and Docker build.