Preloop vs Helicone: Migrate to an actively developed control plane

Preloop is the open-source AI agent control plane: an AI model gateway plus an MCP firewall, per-call human approvals, runtime observability, and a hash-chained audit trail, Apache 2.0 and self-hostable. Helicone is a respected open-source AI gateway and LLM observability platform that was acquired by Mintlify in March 2026. The honest one-liner: if you valued Helicone's open gateway but need an actively developed control plane that also governs tool calls and human approvals, migrate to Preloop.

At a glance

Feature Preloop Helicone
License Apache 2.0 (open source core) Apache 2.0
Self-hosted option Yes, on any infrastructure Yes; on-prem deployment is listed on the Enterprise plan
Development status Actively developed "Helicone's services will remain live for the foreseeable future in maintenance mode": security updates, new models, bug and performance fixes
Primary scope Full control plane LLM gateway and observability
AI model gateway with budgets Yes: OpenAI-, Anthropic- and Gemini-compatible, per-agent attribution Gateway with caching, rate limits, automatic fallbacks
LLM observability Yes: runtime sessions with tool and model calls Yes: strong, mature request and session observability
MCP firewall (tool-call governance) Yes: allow / deny / require-approval / require-justification No
Per-call human approvals Yes: mobile, watch, Slack, Mattermost, email, webhook, CLI No
Audit / evidence Per-account hash chain with signed checkpoints Logs and metrics
Onboards existing local agents Yes: preloop agents discover and preloop agents onboard No equivalent
Pricing Open source free; Cloud and Enterprise plans Hobby free, Pro $79/month, Team $799/month, Enterprise custom, plus usage

What Helicone does well

Helicone earned its adoption. By its own count, users have processed "over 14.2 trillion tokens" through it and "16,000 organizations" have used it (Helicone). It is, architecturally, the closest player to Preloop's gateway.

  • Mature LLM observability. Request logging, sessions, user analytics, cost tracking, and its HQL query language are the reason many teams adopted it.
  • Open source and gateway-first. Apache 2.0, self-hostable, with a clean OpenAI-compatible proxy.
  • Large install base. A lot of teams already route model traffic through it.

The reason to look at a migration is not quality. It is that Helicone's own announcement puts the product in maintenance mode, while the agent governance problem has moved past model-traffic observability. The GitHub repository shows commits continuing at a low pace (the latest on September 16, 2026, at the time of review).

Where Preloop is different

Preloop covers the same model-gateway-plus-observability ground Helicone is known for, then adds the layers an agent control plane needs.

  • The same open, self-hostable gateway, still moving. Preloop's core is Apache 2.0 with an OpenAI-, Anthropic- and Gemini-compatible model gateway, budgets, allowed-model lists, and per-agent attribution.
  • An MCP firewall Helicone does not have. Preloop evaluates every governed tool call against allow / deny / require-approval / require-justification policy on the arguments, not just the tool name. Helicone governs model traffic; it does not govern what an agent does with tools.
  • Per-call human approvals. When a policy routes a tool call for review, a person decides on mobile, Apple Watch, Slack, Mattermost, email, webhook, or the CLI, with the tool, arguments, agent reasoning, and matched policy in front of them.
  • One runtime timeline for tools and models. Runtime sessions show tool calls, model calls, policy decisions, approvals, spend, and outcomes in one place.
  • Audit as evidence. Audit rows are sealed into a per-account hash chain with signed checkpoints: decisions, approvers, justifications, and outcomes you can hand to a security review.
  • Onboarding in two commands. preloop agents discover finds Claude Code, Codex CLI, Cursor, Gemini CLI, OpenClaw, OpenCode, and other agents on a developer's machine, and preloop agents onboard routes the supported ones through Preloop.

Migrating from Helicone

Helicone is OpenAI-compatible and gateway-first, which makes the migration mechanical for the part you already use:

  1. Repoint the gateway. Swap your Helicone gateway base URL for Preloop's model gateway ({PRELOOP_URL}/openai/v1 for OpenAI-format clients, {PRELOOP_URL}/anthropic for Anthropic-format clients). Model traffic, budgets, and per-agent attribution keep working.
  2. Keep your observability. Preloop's runtime sessions cover model-call logging and cost tracking.
  3. Add the layers Helicone never had. Turn on the MCP firewall for tool-call policy and send approvals to the channels your team reads.
  4. Onboard local agents. Run preloop agents discover, then preloop agents onboard, with no SDK changes.

You can run the two side by side during the cutover and retire the Helicone gateway once Preloop carries the traffic.

When to stay on Helicone

  • You only need model-traffic logging and cost analytics, have no tool-call governance or approval requirement, and are comfortable with a product in maintenance mode.
  • Your current Helicone deployment meets your needs and you have no near-term agent governance roadmap.

When to choose Preloop

  • Teams on Helicone who need tool-call governance and human approvals that a model-traffic gateway cannot provide.
  • Platform and DevEx teams rolling out Claude Code, Codex CLI, Cursor, Gemini CLI, OpenClaw, or OpenCode who need policy and approvals, not just observability.
  • Security and compliance teams that need an audit trail that produces evidence for AI Act readiness work.
  • Open-source-first teams that want an actively developed, self-hostable control plane to standardize on.

FAQ

Is Helicone still being developed? Helicone's founders wrote that its services "will remain live for the foreseeable future in maintenance mode", with security updates, new models, and bug and performance fixes continuing. New product capabilities are not promised.

Is Preloop a drop-in replacement for Helicone's gateway? For the model-gateway use case, largely yes. Repointing the base URL keeps model traffic, budgets, and attribution working, and you gain the MCP firewall, approvals, and audit trail. Helicone-specific features such as HQL do not carry over.

Does Preloop keep the observability I relied on in Helicone? Preloop captures model calls and cost in its runtime sessions view and adds tool calls, policy decisions, and approvals to the same timeline.

Is Preloop open source like Helicone? Yes. Both are Apache 2.0. The difference is scope and momentum.

How do I onboard existing agents to Preloop? Run preloop agents discover to find local agent configurations, then preloop agents onboard to route the supported ones through Preloop. Onboarding backs up the existing config, and preloop agents restore puts it back.

Try Preloop

Last reviewed: 2026-10-08.