March 2026: Competitive Response & Platform Maturity

Published March 14, 2026

This was a large cross-cutting release spanning 155+ files and 13,000+ lines of code. It was driven by competitive shifts — Helicone was acquired by Mintlify, Langfuse was absorbed by ClickHouse, and the remaining players (Portkey, E2B, Daytona) raised significant rounds. Our response: ship the integrated stack that nobody else offers.

OpenClaw v2026.3.13 Upgrade

We upgraded from v2026.2.26 to v2026.3.13 across the entire platform — 81 files, 371 replacements, and 3 breaking changes handled.

Key changes in the new OpenClaw version:

• Gateway auth now requires explicit gateway.auth.mode
• Tools profile defaults to messaging
• ACP dispatch enabled by default
• Claude 4.6 adaptive thinking support
• Docker multi-stage minimal images

51 LLM Providers (was 18)

The gateway now routes to 51 providers across 7 tiers:

Every provider gets the same governance chain — rate limits, cost tracking, PII scanning, model allowlists, and HITL approvals.

Gateway Response Cache

A two-tier caching system that eliminates duplicate LLM calls:

• Exact-match caching — SHA-256 hash of normalized request body (model, messages, temperature, tools)
• Per-org cache policies — configurable TTL (60s to 24h), per-model exclusion lists
• Cache statistics — hit rate, total entries, memory usage via admin API
• Streaming bypass — streaming requests (stream: true) skip the cache automatically

Enable via the FF_GATEWAY_RESPONSE_CACHE feature flag, then configure per-org in the dashboard.

Read the full cache documentation →

EU AI Act Compliance Engine

Automated compliance scoring against 8 EU AI Act articles for high-risk AI systems:

Each article gets a 0-100 score with compliant, partial, or non_compliant status. One-click remediation fixes common gaps (enable PII scanning, set HITL thresholds, configure budgets).

Audit trail exports in JSON or CSV with chain-of-custody metadata for regulators.

Read the full compliance documentation →

Orchestration Patterns

Three governance-aware multi-agent orchestration patterns for managed runners:

Sequential Pipeline

Ordered agent stages with per-stage cost budgets, output chaining, and automatic rollback on governance denial.

Parallel Fan-Out

Concurrent agent execution with quorum voting (ALL, MAJORITY, N_OF_M, FIRST strategies), weighted branches, and optional synthesis.

Hierarchical Delegation

Manager agent delegates to specialists with depth limiting, transitive governance, and per-specialist cost tracking.

All patterns route LLM calls through the gateway, so the full governance chain applies to every call.

Read the full orchestration documentation →

Open-Source Governance Chain

We extracted the core governance logic into a standalone MIT-licensed package — 7 files, 732 lines.

Available at packages/governance/ for teams that want to self-host the policy pipeline without the full platform.

Firecracker Stability

• Health monitoring with jailer and prewarm lifecycle management
• Provisioning selector for optimal allocation
• 30 integration tests covering the full runner lifecycle

Rate Limit Headers

Enhanced 429 responses now include structured rate limit info:

HTTP/1.1 429 Too Many Requests
X-RateLimit-Limit: 60
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 1710432000
Retry-After: 30

SDK Parity

Python and TypeScript SDKs now cover snapshots, A2A communication, compliance, and CLI commands — 9 new files across both SDKs.

SEO Comparison Pages

Four new comparison pages on the dashboard:

• Curate-Me vs Helicone (acquired by Mintlify)
• Curate-Me vs Langfuse (acquired by ClickHouse)
• Curate-Me vs Portkey
• Curate-Me vs LiteLLM

What’s Next

• Design partner onboarding — 10 target teams in the pipeline
• ClawHub Security Scanner — free viral tool for the community
• SSO/SAML — enterprise authentication
• Multi-region runners — cross-region warm pools on Hetzner

Stay up to date by following the changelog or emailing hello@curate-me.ai.