heka

Software factory consulting · Since 1999

We build the software factory.

Companies no longer pay engineers to type code. Agents write it. We are paid for decades of shipping software, applied to the system that ships: the harness around your agents.

Start with an audit How we work
Guides AGENTS.md · rules · skills · specs
Agents dozens to hundreds, one codebase
Sensors tests · QA · holdouts · review · gates
Production visible · reversible · boring
Fig. 1 The harness: guides steer the agents, sensors check their work, and only what passes ships.

01 The shift

Consulting changed.

Then

Writing software was expensive, so companies hired engineers to write it. Consultants sold hours at a keyboard, and the deliverable was code.

Now

Agents write the code. The expensive part is everything around them: telling them what to build and how, checking that what they built is right, keeping secrets out of their hands, and keeping production steady while the number of changes goes up a hundredfold.

That is the work now. We build the factory, and we leave the lights on: every change visible, verified and reversible.

02 What we build

Five parts of a factory that ships.

  1. 01 Guides

    A codebase agents can read

    Layout on disk, AGENTS.md and CLAUDE.md, rules, skills and specs: what an agent reads before it writes a line.

  2. 02 Sensors

    Verification at volume

    Tests, QA pipelines, holdout scenarios, review agents and merge gates that can check hundreds or thousands of changes a week.

  3. 03 Capabilities, not credentials

    Access without exposure

    Agents get capabilities, never credentials. Vault-brokered secrets, scoped MCP servers, least privilege, audit logs.

  4. 04 Orchestration

    Many agents, one codebase

    Parallel agents in isolated worktrees, agent hosts, MCP servers, skills, triggers and scheduled jobs: dozens to hundreds of agents on one repo.

  5. 05 Infrastructure

    Reliability at agent speed

    Progressive delivery, rollback and observability, so production at thousands of changes a week is as steady as it was with weekly deploys.

  6. →Where to start

    The audit covers all five.

    One to two weeks, fixed fee: written findings and the first commits.

03 Proof

We run our own factory.

Everything we build for clients, we run ourselves first. heka's own work goes through the same harness we sell.

  • Several agent hosts Agents work in parallel on dedicated machines, each change in its own worktree.
  • Heka Vault A self-hosted secrets vault. Agents and pipelines get scoped keys; credentials stay out of every context window.
  • Self-hosted forge and CI Every change is a pull request, and every pull request runs the same gate locally and in CI before it merges.
  • Per-project notifications Each project reports its builds, deploys and failures to its own channel.

04 Engagements

Three ways to start.

What each one includes →

05 Writing

Notes from the floor.

→Lights on

Start with an audit.

Tell us what you ship, how many people and agents work on it, and what keeps breaking. Eric reads and answers every one himself.

Email us