mywavelength.ai
Wavelength
Standalone product
Coaches you to write to the human on the other end, not at them. Slack and Gmail suggestions adapt to the recipient without storing message content.
Next.jsNestJSSlackGmailClaude
Visit mywavelength.ai →freelance agent infra · powell river, bc
I build agent infrastructure you can actually run.
Sonder, Engram, Ginnung — the stack. Wavelength, DripCtl — the products that use it, so the stack stays honest.
One-person studio in Powell River, BC. Some days that means agent infrastructure, some days a Salesforce integration or a product that started as a weird note and refused to stay quiet.
Read the writing →
Currently shipping: Aegis Phase-0 · Engram 78% on LongMemEval · Wavelength Slack live
Project directory
One root domain, multiple product worlds.
The root stays neutral. Project sites can use their own palette, layout, voice, and conversion goals without pretending they belong to the same product family.
Products
In progress
dripctl.dev
DripCtl
Developer tool
Lifecycle email you write in TypeScript instead of a drag-and-drop builder. Send through your own provider, let Cadence tune the timing.
NestJSBullMQPostgreSQLTypeScript
Visit dripctl.dev →thetrashpanda.ca
The Trash Panda
Experiment
A scrappy product lane for the useful, weird, half-feral ideas that do not belong inside the agent stack but still deserve daylight.
PrototypeBrandAutomation
Visit thetrashpanda.ca →chalkbreaker.com
ChalkBreaker
Founder
Runs a few million playoff brackets so you don't have to. EV-optimized picks, no vibes, every claim points back to the data.
Next.jsTypeScriptVercelStripe
Visit chalkbreaker.com →Agent Stack
ginnung.ai
Ginnung
Control plane
The cockpit on top of Sonder. Wire the faculties, watch them run, and see what the agents are actually doing.
Next.jsControl PlaneAgents
Visit ginnung.ai →sonder.ginnung.ai
Sonder
Runtime
A signed receipt for everything an agent does. Memory, reasoning, governance, capability, prediction, and intent all get logged as auditable events you can replay.
TypeScriptSQLiteed25519SDK
Visit sonder.ginnung.ai →engram.ginnung.ai
Engram
Memory faculty
Memory for agents that survives the restart. Vector plus structured storage, confidence scoring, and a dream cycle that consolidates while nobody's watching.
NestJSPostgreSQLpgvectorTypeScript
Visit engram.ginnung.ai →acr.ginnung.ai
ACR
Capability faculty
Agent Capability Runtime. Loads the right tools at the right resolution so the agent isn't drowning in context it doesn't need.
TypeScriptContextCapabilities
Visit acr.ginnung.ai →parliament.ginnung.ai
Parliament
Reasoning faculty
Multi-model synthetic discourse. Runs consensus and adversarial critique across models so one agent's blind spot isn't the whole answer.
TypeScriptMulti-modelConsensus
Visit parliament.ginnung.ai →lattice.ginnung.ai
Lattice
Governance faculty
Coordination infrastructure and gate policy for multi-agent systems. Decides who can do what, when, and with what blast radius.
TypeScriptPolicyCoordination
Visit lattice.ginnung.ai →awm.ginnung.ai
AWM
Intent faculty
Looks ahead in an agent pipeline so it fails less and costs less. Planned-action records plus routing signals do most of the work.
TypeScriptSignalsRouting
Visit awm.ginnung.ai →Labs
github.com/heybeaux/inos
Inos
Reasoning surface
A 4D reasoning graph for humans and agents to think together. A shared canvas for argument structure, evidence, and decisions.
TypeScriptGraphReasoning
View on GitHub →github.com/heybeaux/tra
TRA
Research analyst
A deliberated dev/AI research analyst. Gathers evidence, argues both sides, and returns a backed recommendation instead of vibes.
TypeScriptAgentsResearch
View on GitHub →github.com/heybeaux/aegis
Aegis
Governance harness
Predictive OSS governance harness. Catches risky agent behavior before it ships, with a tool-use regression floor as the north star.
TypeScriptEvalGovernance
View on GitHub →Client Work
generositycatalyst.com
Generosity Catalyst
CTO
Lightweight CRM for nonprofits that want real donor relationships, not automation funnels. Architecture, infrastructure, and weekly feature shipping.
Next.jsSupabaseStripe
Visit generositycatalyst.com →whalehawk.com
WhaleHawk
Full-stack developer
Sales intelligence for healthcare. Landing site, web app, and Salesforce integration that brings prospect intelligence into the CRM.
ApexLWCReact
Visit whalehawk.com →wildernesscommittee.org
Wilderness Committee
Digital fundraising coordinator
Salesforce admin, development, automation, and integrations during a major migration. Helped drive 40% fewer lost monthly donors over two years.
SalesforcePardotShopifyFormAssembly
Visit wildernesscommittee.org →What heybeaux is
The common thread is systems thinking: find the moving parts, make the relationships visible, and build the smallest thing that turns confusion into momentum. In the agent stack, that means Sonder, Ginnung, and the six faculties. In standalone products, it means letting each project have the shape its audience needs.
The root domain should make that range easier to understand, not flatten every project into the same visual language.
Writing
Recent essays and notes.
Mar 20, 2026
How I Used Karpathy's Autoresearch to Grade-A My AI Stack
One optimization loop. Three systems. Every one improved. The method is stupid simple and works on anything you can measure.
Feb 20, 2026
What Happens When You Give AI Memory, Then Identity, Then Awareness
A devlog about building Engram — and the two AI agents who helped build it. Three wipes, 70 tickets, and the line between tool and teammate.
Feb 4, 2026
Teaching AI to Remember When
I asked my AI what we did today. It had no idea. So we taught it about time — and in the process, discovered why memory is harder than anyone thinks.
Feb 3, 2026
The Bug That AI Couldn't See
Four hours. Two humans. One AI. A simple form that wouldn't work. Sometimes the answer isn't in the logic — it's in the gap between what should happen and what actually does.
How I work
Four commitments.
The shape follows the project.
A control plane, a communication coach, an email engine, and client delivery work do not need the same visual language. The root domain gives them room to be distinct.
Useful beats theatrical.
The best systems make the work clearer. Strip out the ceremony, keep the evidence, and ship the thing someone can actually use.
The audit trail still matters.
For agent infrastructure, the record is the product. Memory, reasoning, governance, capability, intent, and prediction belong in a trace someone can inspect.
Small teams can still build seriously.
A practical generalist with good tools can move fast without making a mess. That is the operating assumption behind heybeaux.
Work with me
Building something that needs a practical generalist?
Send the messy version. I can help shape product direction, ship the code, wire the systems together, or figure out why the thing nobody owns keeps breaking.