LimitedGet $1,000 of creditsBook a demo →

Transform your team’s knowledge into code

Describe your work in plain English. hegl writes real software for it, tests it against your data, and gets better every time you correct it.

Leading teams are using hegl to turn their organizational judgment into code.

Grab
datajolt
ozow
Enhancv
renofi
hygraph
mize
stay22
“With hegl we were able to turn over 90% user fixes into permanent rules so the same bugs in parsing resumes stopped coming back”
Alex GinovskiHead of Product @ Enhancv

Why you still don’t trust AI to automate anything that actually matters

AI works in software because you can check if the code compiles and billions of examples exist. Most of the work out there doesn’t have that. The judgment the work depends on was never written down, so there’s nothing to test against and nothing to learn from. SOPs and existing documentation are enough to produce a demo, but proves insufficient to ship a reliable system.

An iceberg, most of its mass below the waterline
What's written✓DOCUMENTS✓RULES✓SOPs✕ Changes✕ Nuance✕ Trade-offs✕ Variance✕ Dependencies✕ Judgment✕ Context✕ Risks✕ ExperienceVisibleInvisible

THE LOOP

Runs the work. Captures the judgment. Rewrites itself.

Discover

Tell hegl about the workflow: type, talk, drop in your SOP, share examples. It asks in-depth questions, prompts you to connect the systems involved, and builds a picture of what done right looks like.

[inputs: connected]state: alignedraw_workflow_datasop_structure_verified

Decompose

The workflow becomes deterministic software. The genuine judgment calls are isolated into small decision points with fixed inputs and outputs – everything else runs the same way every time until you change it.

[ monolithic ]raw_flowunstructuredlogic_blockstep_01: validatedeterministic [pass/fail]step_02: executedeterministic [automated]decision_box: judgmentfixed_in ➔ human_eval ➔ fixed_outstable_output[ repeatable ]output_02 [REJ][ISOLATED DECISION POINT]

Build & Test

hegl builds the system, assembles an eval set from your real production data, and tests itself against it until it hits success criteria. It goes live when you decide to publish it.

ASSEMBLYprod_eval_setn = 500 actual casesschema_check: latency_bench: edge_cases: Decision: User/AutoTarget MetReady8 iterations · last evaluated 2 mins ago

Compound

Every correction becomes a permanent test, and the system rewrites itself to pass it. The eval set grows with every override, feedback or change of context - accuracy compounds over time instead of drifting.

HEGL :: COMPOUND_ENGINE|eval_set_v4.12|mode: auto_rewrite [active]n=4,890 testsINGESTION STREAM+ Human Overrideid: ovr_9012 [saved]+ Context Changeschema: v2.4_patch+ Edge Case Logexception: routing_null3 New Tests ReadyInjecting to suite ➔PERMANENT EVAL SUITEtest_suite.size: 4,890> assert_eval(override_9012) == PERMANENTtest_001..1200 [Baseline]PASSEDtest_1201..4887 [Historical]PASSED+ test_4888 [ovr_9012]Auto-rewritten system logic appliedLOCKEDSystem Self-Rewrite Coverage100.0%Zero unhandled edge cases recorded across 4k+ runsPERMANENT EVAL SUITE99.8%↗ COMPOUNDING↘ Standard Drift (84%)T_0 (Deploy)T_FINAL ↗✓ Zero Drift Active

The product

Start on real files today and let it run alongside your team.

Elicits what was never written down.

hegl asks questions about the workflow and turns the answers into a detailed spec.

Real evaluation set

Built from files your team actually worked. Nothing ships until it passes on them.

[Agent has a question before it cancontinue building]For ambiguous leads, should I default thesource to [Lead.LeadSource] or...[43 Updates on the agent flow arecompleted]Completion score has increased from96.4% > 98.2% on the last run[Should I consider the March bonuspayment in the income calculationfor the Ferguson file?]Paused for a person to approve or confirm

Asks when it doesn’t know

And the answer becomes a permanent case in the eval set.

In progressBuilding v5Cancel RunVersion historyv4CandidateYesterdayPublishes straight to an artifactv3Published1 week agoPre-v1 experiment. Single-box monolithic prompt;v2Previous1 month agoThe v2 market_briefer adds market regime, top moversv1Previous3 weeks agoWires the v1 market_briefer — index levels in plain-textv0.9Archive1 month agoAdds a per-sector breakdown on top under build

Rewrites itself

New evidence, new version, re-run against the full eval set. Every version beats the last.

calc_qualifying_incomeReadyRead the paystubs, W-2s and VOE the way the processor does.Work out what income actually qualifies.MATCH TO OPERATOR0.94Used in:income-review-v1 +2 moreNot published11 iterations · last about 2 weeks ago

Judgment isolated

LLMs decide only inside schema-fixed decision points. Everything around them runs the same way every time.

fr-01KY8KN5P2ARC6XVVDPNH3CM4PPausedStarted 23 Jul 23:06 · Open flow definition (v4)Search this run...[Flow started]Run input

Auditable by construction

Append-only record of inputs, decisions, overrides, versions.

The product

In an AI first world, your team’s knowledge is your biggest asset

Automation is the yield. What you own at the end is a machine-readable record of how your organization decides - the layer no vendor can sell you and no model already has.

Everyday patterns: Read incoming documents, enter what matters, Review requests, approve or reject, Check submissions, chase what’s missing, Compare sources, resolve mismatches, Answer tickets like your best person, Decide each exception, then do it, Check contracts against playbooks, Fill forms from documents, Match invoices to orders, Triage and route the inbox

Use cases

Built for the work
you actually have

hegl is built to work across many domains where information comes in and a person decides what happens next. Click on the folders below to explore some of the common use cases.

Book a demo

Getting Started

Build, test and deploy your first agent in a single day, not weeks.

Enhancv
I built an automation that handles refund tickets in an afternoon without pulling engineers off the roadmap. What used to be handled by 1 FTE is now done in a couple of minutes.
– Volen, Head of Product @ Enhancv

80%+

of refund requests
auto-resolved

30h+

Saved every month
for the team

  • Triage new tickets
  • Automated refunds
  • Employer matching
  • Invoice reconciliation
  • Automated contract review
  • New employee onboarding
  • Bug reporting
  • Dispute handling

The reasoning that runs your organization
should be an asset you own.

Book a demo