The reliability layer for agentic AI
Ship reliable agents that automate manual workflows.
Built in an afternoon.Corrected in plain language by the people who know the work.Rewritten and retested by our harness, not by you.
The problem
Building a v1 is quick. Getting it to prod is hard. Keeping it running is a never-ending loop.
Every automation an engineer owns turns into this cycle, and it never ends.
How it works
You build it once. Everyone else keeps it working.
You
Describe the process, connect the systems, publish. An afternoon, from the harness you already use. Then you are on to the next problem.
hegl
Turns the process into a deterministic flow, isolates judgment into decision boxes with their own evals, builds the eval set from real cases, and runs the autoresearch loop that rewrites and retests it for as long as it runs.
The people who know the work
Underwriting, support, ops: whoever does it by hand today. From week one they correct the agent in plain language, tweak the implementation in chat, and ask it questions. Every correction becomes a permanent test.
For developers
The workflow engine is the easy half.
Pricing
Pay for what you use.
Bring your own MCP client for $9 a month, or use the entire platform for $299. Credits cover the rest.
Bring your own harness
Claude Code, Codex, Cursor or any MCP client
$9
per month, plus credits
Your plugin builds and rebuilds the agent on your machine. Credits pay for runs only.
For an engineer automating one process for their own team.
- Plugin for your harness
- Hosted flows, decision boxes and traces
- Evals and autoresearch, run by the plugin
- 500 flow runs a month included
Platform
Everything, in the app
$299
per month, $100 in credits included
Build in the app from a discovery conversation. Credits pay for builds, rebuilds and runs.
For a team handing the automation to the people who run the process.
- Everything in Bring your own harness
- Discovery conversation builder
- Hosted eval sets and autoresearch
- Decision boxes trained on corrections
- Feedback and error tracking per run
- Operator chat: fix issues, ask questions
- 5,000 flow runs a month included
Enterprise
For the whole operation
Custom
committed credits
Credits at volume, SSO, audit exports, a custom DPA and SLA, private deployment.
For an operations org running many agents under one contract.
- Committed credit volume
- Flow runs at volume pricing
- SSO and audit exports
- Custom DPA and SLA
- Private deployment
- What are credits?
- Credits pay for usage. On Bring your own harness that is runs. On Platform it is builds, rebuilds and runs. Buy them as you go in the app.
- What is a flow run?
- One execution of a deployed flow, with or without a model call. Each tier includes a monthly allowance; beyond it, a run is 0.01 credits.
- Which harnesses work?
- Claude Code, Codex, Cursor, or any MCP client. The plugin is the same; the flow it builds is the same.
- Where do my evals live?
- On the plugin tier, on your machine, in your repo. On Platform, hosted in the app, where autoresearch keeps them current.
- What does Platform add?
- The discovery builder, hosted evals, decision boxes trained on corrections, feedback tracking, and the operator chat.
- Can I move from one to the other?
- Yes. The agent, its evals and its history come with you.
Building something real? We cover the credits.
Apply for an inference grant and run on us while you build.
For teams automating their own operations, open source projects, and developers building in public.
A short form, no deck. We reply with a decision.
Ship reliable agents.
From your favourite harness, any MCP client, or the hegl app.