Obsidian Labs · AI automation & consulting · est. 2023

Agents that do the work.Not chatbots that talk about it.

Obsidian Labs is an AI automation and consulting studio for operating businesses. We put agents inside the inbox, the CRM and the dispatch desk — reading, replying, quoting, booking, generating the paperwork — and hand the thread to a person the moment it matters. You own everything we build.

StatusSystems in production
FitOperating businesses, roughly $10–50M, with a real workflow to run
EngagementAudit → Build Sprint → Run
Crown +1.52 · base −1.62 · taper 0.68 · ring ×5 · grid 24Fig. 01 — system schematicProductionInbox →→ HumanAgent01 — AI Automationagents in the inbox, CRM, dispatch desk02 — AI Strategy & Consultingaudit · roadmap · fractional AI lead03 — Custom AI Systemswhen off-the-shelf won't do
Fig. 01 — struck, not polished
01Services

Automation first.
Custom when it has to be.

Most of what an operator needs is an agent inside a tool they already run, wired properly. We start there. When off-the-shelf genuinely won't do, we build the system.

01

AI Automation

Agents inside the tools you already run. They read, reply, qualify, book, quote and generate the paperwork, and escalate to a person on rules you set.

  • Inbox agentstriage, drafting, RFQ round-trips over plain email
  • CRM & SMS agentsqualification, booking, follow-up with memory that spans days
  • Workflow automationcron fleets, inbox routing, documents generated at the moment of action
  • Handoff rulespricing, complaints, anything that matters goes to a human
InboxCRMDispatchDocuments
Automation
02

AI Strategy & Consulting

Know what to automate before you automate it. An audit maps your workflows, ranks them by leverage and risk, and comes back as a written roadmap.

  • Automation auditworkflow map, ranked opportunities, written findings
  • Roadmapsequencing, build-vs-buy, data and integration prerequisites
  • Fractional AI leada builder in your leadership loop, part-time
  • Team enablementtraining on the systems you now own
AuditRoadmapFractionalEnablement
Strategy & Consulting
03

Custom AI Systems

When off-the-shelf won't do. A freight desk that books over email, a CRM whose agents are staff — designed and shipped end to end, and you get the keys.

  • Agentic platformsagents with tools, memory and guardrails, embedded in the software they operate
  • IntegrationsERPs, telephony, email, auth, payments
  • Document enginesrate cons, BOLs, PDFs generated when a load is awarded
  • Voice & messagingsoftphones, SMS at conversation depth
Next.jsSupabaseClaudeInngest
Custom AI Systems
02Engagements

Start with an audit.
Build only what it justifies.

Three named engagements. Each one stands alone, ends with something you keep, and is the only prerequisite for the next.

01

Automation Audit

Fixed scope · written output

We sit with the people doing the work, map the workflows, and rank what an agent could take over — by leverage, by risk, and by what your data can support today.

OutputA written roadmap: what to automate, in what order, and what it depends on. Yours either way.
02

Build Sprint

Fixed scope · one workflow live

One workflow, end to end, in production. Not a sandbox — the actual agent doing one real job with real data, early enough to change course cheaply. Then shipped increments, weekly, in readable PRs.

OutputOne real workflow running in production, in your repo.
03

Run

Retainer or fractional

We stay on to run, extend and harden what shipped, or as a fractional AI lead in your leadership loop. Month to month. Everything keeps running if we leave.

OutputSystems that keep improving, and a team that can run them.
Scope first, then a number. No prices on this page on purpose.How we work
03Work

Proof, running
in production.

Systems we designed, built and run for operating businesses. Real freight, real conversations, real paperwork. Not prototypes, not decks.

Case 01In daily production

Biomass Buddy

Freight booking OS — built for a working biomass supplier

A dispatch desk that books truckload freight the way a good broker would — over plain email. It sends RFQs to the right carriers, reads their replies, captures a number in a sentence as a structured bid, and generates the rate con, BOL and packing slip the moment a load is awarded. The dispatcher stays in one inbox; the system does the legwork.

  • AI dispatch deskcarrier RFQ automation over email and SMS, bid capture from free-text replies
  • Live team inboxa mailbox agents and humans share, with notes, mentions and routing
  • Full TMSorders to dispatch to delivered, with generated rate cons, BOLs and packing slips
  • Carrier intelligencescoring from all-time lane history, plus in-browser calling
Next.jsSupabaseClaudeTwilioResendInngest
Read the engagement
Buddy Ops — dispatch desk
Sample threadProduction
RFQ sentwalking floor · this lane · carriers on file
to: carriers · from: dispatch@ · subject: Load BB-1047 — Thu pickup
Two loads of mulch, walking floor, 23.5 t each. Thursday pickup, Friday delivery. What’s your all-in rate?
Carrier repliedfree text · read by the agent
from: dispatch @ carrier · re: Load BB-1047
Morning — we can cover that Thursday. $1,850 all in, driver’s already in the area.
Bid captured from the replystructured · no re-keying
QuoteQ-218 · $1,850 all inLaneKnown carrier · history on fileStatusAwaiting award
Awarded → paperwork generatedat the moment of action
RC-0417 BOL-1047 PS-1047
Case 02In active development, in production

Atlas HQ

AI-powered CRM & marketing suite — running live client conversations

A CRM and marketing platform where the agents are staff. Ana, its SMS agent, works leads with memory that spans days — answering, qualifying, booking — and hands the thread to a human the moment it matters. Around her: pipelines, campaigns, intent-lead sourcing with enrichment, and a release train that ships to production three mornings a week with client release notes drafted automatically.

  • SMS agents with memorylong-running conversations, qualification, booking
  • Marketing suitecampaigns, pipelines and contacts with agent legwork built in
  • Lead sourcingintent data, enrichment and verification before a human dials
  • Release trainMon/Wed/Fri production deploys, notes drafted automatically
Next.jsSupabaseClaudeTwilioExpoVercel
Read the engagement
Atlas HQ — conversations
Sample threadProduction
Lead · 4:12 pmHi, saw your ad. Do you do gutter guards too or just roofs?
Ana · agent · 4:12 pmWe do both — guards usually go on with a roof job or on their own. Want me to set up a quick look at the house this week?
Lead · Thu 9:03 amSorry, missed this. Thursday afternoon works.
Ana · agent · Thu 9:04 amNo problem. Thursday 2:40 is open — I’ll hold it and text you the confirmation.
BookedThu 2:40 — synced to pipeline
MemoryThread resumed after 2 days, no re-ask
HandoffPricing, complaints, anything that matters → human
Release trainMon 06:00 deployed · notes drafted
Wed 06:00 deployed
Fri 06:00 assembling
Case 03

Mittis

Telecom infrastructure · 2025–26

Carrier-grade telecom underneath the platforms above — numbers, 10DLC registration and compliance, programmable voice, and a softphone that rings the right person. The calling layer inside the systems we build.

In active development, in production
+From the lab — we are our own first client
InternalJarvis

Our in-house assistant: email, calendar and approvals, by voice and by message. The inbox agent we use before we sell one.

InternalSecond Brain

A knowledge hub every AI surface we run can read and write. Studio memory as infrastructure.

LiveDebtPilot

An AI debt-payoff coach: plan, payoff order, and a coach that knows your numbers.

FoundationKeelstone

A multi-tenant compliance platform foundation — certificates, audits, and row-level security from day one.

04Principles

Obsidian forms when something molten cools fast — and it holds an edge finer than steel. That’s the software we make: formed under real pressure, sharpened on real work.

P-01

Production is the only demo.

We don't present mockups of what an agent might do. We put it in your inbox, doing it.

P-02

We are our own first client.

Our inbox agent, our telephony, our knowledge hub run on the stack we recommend. Nothing ships to you that we don't run ourselves.

P-03

AI-native, not AI-flavored.

Agents live inside the workflow — reading the inbox, working the lead, booking the truck — not in a widget in the corner.

P-04

Owned, not rented.

Your repo, your infrastructure, your data. If we disappeared tomorrow, everything would keep running.

Studio, not SaaS. The people on the call are the people who build.

About the studio
Stack
  • Next.js
  • React
  • TypeScript
  • Supabase
  • Postgres
  • Claude
  • Twilio
  • Telnyx
  • Resend
  • Clerk
  • Inngest
  • Expo
  • Vercel

Book an audit.

Thirty minutes on a call first. We'll tell you what we'd automate, in what order, and what it depends on — in writing, whether or not you hire us.

Studio, not SaaS — no pricing page, no free trial, no sales team.