01 Select: parts and BOM
Distributor searches and spec comparisons against datasheets, stock and flight heritage.
- Part search
- Spec comparison
- Heritage lookup
MHz Labs builds RF power electronics for plasma thrusters. A small founding team does the engineering; The Hive, our agent team, takes the research, search and simulation load.
hive.dispatch(card)
conductor · 18 cards queued
4 roles · 18 cards in flight · 0 conflicts
Where agents help today
Every run has a conductor, a shared board, hard loop limits and an engineer who signs off. Agents extend a small team; engineers make the calls.
One agent splits the goal, owns the board and assigns cards. When its context fills, a fresh conductor walks on and reads the same state.
State lives on disk, not in anyone's memory: goal, decisions, cards and a log every agent reads before it acts.
Datasheets, standards, programmes and prospects, gathered with sources and flagged when unverified.
Parallel cards in separate files: part searches, scripts, simulations and drafts. One owner per card, no overwrites.
Three attempts per card, then a written meeting with a time limit. Every dispute ends in a recorded decision.
Nothing outward facing leaves on an agent's word: no email, order or release goes out without a founder.
A run starts with one state file: the goal, the definition of done, who does what and what nobody may do. Every agent reads it first, and every decision is written back to it.
hive.yaml
# hive.yaml
run: rf-ppu-pa-board
state: [goal, board, roster, log]
conductor: { reads: state, rotate_at: 65% }
players:
ladder: [fast, standard, deep]
escalate: on_failed_check
owner: one_per_card
cards:
verify: artifact_not_report
limit: { attempts: 3, then: meeting }
guardrails:
outward_actions: founder_signoff
883
flight-heritage COTS parts catalogued with agents
4
roles in every run, one conductor
3
attempts per card, then a meeting
24h
a day the hive can run, night work ready by morning
The conductor splits the goal into cards small enough for one agent, each with inputs and acceptance criteria.
Researchers, workers and writers take cards in parallel and write results to a shared cache.
A reviewer checks the artifact itself, not the report. Then an engineer signs off.
Agents do the legwork around the board: they search distributors, compare specs, keep the BOM and draft reference circuits from datasheets. Engineers wire the schematics and route every board by hand.
Distributor searches and spec comparisons against datasheets, stock and flight heritage.
Component groups and first-pass reference circuits drafted from the datasheet; an engineer wires the sheet.
Placement and routing done by hand. Agents run the thermal and field simulations on the result.
The hive's toolbox
Trademarks belong to their owners. No endorsement implied.
Agents map propulsion programmes, OEMs and the engineers behind them, with a source for every row.
First drafts written to each engineer's own work, sent only after a founder has read them.
Standards, competitors and funding programmes, read and cited while the team sleeps.
Applications, proposals and decks assembled from one source of truth.
● In design
Next, the hive gets hands. We are designing a flying probe tester: two probes on XY carriages that land on a board's test pads, measure, and hand the result back to the design.
Move over the board to probe it.
Building an RF thruster? We move at the speed of a hive. Talk to the engineers behind it.
This page is illustrative. Our agent orchestration is not.
The cursors are an animation. The conductor, shared board, attempt limits and founder sign-off are how we really work: agents extend our manpower.