Products
Trace every LLM call, costs, and hallucinations
HIPAA, FDCPA, GDPR enforced on every LLM call
Managed memory and RAG with citation enforcement
Version-controlled prompts with A/B testing and rollback
Memory-aware AI assistant for any website
Two infrastructure layers every serious agent team ends up rewriting from scratch: observability (Peekr) and memory(Extremis). MIT-licensed libraries underneath, drop-in hosted services on top. Built for Series A–C teams shipping AI in production.
The products
Each product solves one hard problem AI teams keep rebuilding from scratch. Pick one, stack all three, or self-host the open-source core.
See exactly what your agent is doing — and what it's costing.
Two lines of code. Auto-instruments OpenAI, Anthropic, Gemini, and Bedrock. Every LLM call, tool use, and hallucination score in a shared dashboard — no proxy, no rewrite, no framework lock-in.
Try it free →
Managed memory and RAG with citation enforcement.
Your agent answers from your own data — verified, cited, and updated as you go. Every recall is grounded. Wrong answers get flagged before they reach production.
Try it free →
A memory-aware AI assistant for any website.
One script tag. A Friday widget on your site that already knows your content and remembers returning visitors. Built on Extremis + Peekr.
Join waitlist →
No lock-in by design
Every managed service is an MIT-licensed library underneath. If we ever stop shipping — or you decide to bring everything in-house — your code keeps running on the exact same engine. No version skew, no migration, no leverage we hold over you.
Knowledge base engine · MIT
Python library + FastAPI server + TS SDK. The engine Extremis Cloud is built on. SQLite by default, Postgres + pgvector for scale.
Design partnership
For a small number of customers each quarter, we run a paid design partnership — our engineers and yours co-build the integration end-to-end, ending with the product live in your stack and a shared technical write-up.
Shape
Fixed scope, fixed fee. Typical engagement is 3–6 weeks ending with the integration deployed, the eval harness wired, and a runbook your team owns.
Investment
Engagement fee covers the build. A 12-month subscription to whichever Cloud product we deploy is bundled in. No ongoing services retainer.
Capacity
We run one design partnership at a time so the product work doesn't suffer. If we're booked, we'll tell you the next opening and the self-serve path until then.
What our products are built for
Visitors return weeks later and the agent picks up where it left off — plan, last ticket, features tried. Refund policies stay accurate because they live as structured memories, not in a stale prompt.
Procedural memories capture how your team runs incidents, hires, ships. New engineers ask the copilot instead of pinging seniors. Wrong answers get flagged at write time, not in production.
Every recall returns doc + chunk + byte-span. Your support bot stops hallucinating because the retrieval layer refuses to surface ungrounded answers — and the dashboard shows you the ones it caught.
Episodic memories track every PR review and bug fix. Semantic memories distil into 'how this team does X.' Cursor-style copilot, but it adapts to your team's conventions instead of guessing.
How we build
01
Every Cloud product is an MIT-licensed library underneath. Read the source, run it on a laptop, self-host in your VPC. The hosted tier is a convenience, not a moat.
02
Stateless RAG bots are demos. Real products need layered memory with verification, citations, and a feedback loop. We treat it as platform, not as a feature.
03
Agent stacks don't fail at low traffic — they fail when nobody can see why an answer changed last Tuesday. We ship the traces before we ship the optimisation.
04
SOC 2 Type 1 in flight, self-hosted edition documented today, SSO + audit logs landing through Q3. We tell you exactly what's ready and what isn't — no roadmap theatre.
Who builds this
StarkSphereLabs builds Peekr and Extremis after watching the same two problems — observability and memory — get rebuilt from scratch on every production AI project. The OSS libraries came first; the hosted tiers exist for teams that want the same engine without the ops overhead.
Open source
MIT license, read every line
10+ compliance packs
HIPAA, FDCPA, FINRA, GDPR, EU AI Act…
SOC 2 Type 1
Audit in flight, targeted Q3 2026
Free tier
10k spans/month, no card needed
Join the Peekr community
FAQ
AI infrastructure for engineering teams shipping agents. Two live products: Extremis Cloud (managed memory + RAG with citation enforcement) and Peekr Cloud (LLM and agent observability). Both are MIT-licensed underneath — run the OSS yourself, or use the hosted tier. Drop-in Friday (the website agent) is in build and lands next.
Series A–C engineering teams shipping AI in production. Typical buyer is a CTO, VP Engineering, or ML platform lead at a 20–200 person company. If you're at pre-seed or true F500, we're probably not the right fit yet — talk to us anyway, but we'll be upfront about it.
Both. Extremis and Peekr are MIT-licensed — run them on a laptop, in your VPC, anywhere. Extremis Cloud and Peekr Cloud are the same engines, fully managed. Identical APIs, no version skew, no lock-in. Switch direction any time without rewriting.
Honest answer: SOC 2 Type 1 audit is in flight, report targeted by end of Q3. Until then, the self-hosted edition is the answer to compliance — same MIT engine, runs entirely in your VPC, no data leaves your network. SSO/SAML, audit logs, and RBAC land alongside Type 1 in Q3.
Only as an inbound design partnership. We run one paid engagement at a time (typical scope $15k+, 3–6 weeks) where our engineers co-build the integration with your team. It's outcome-priced, not hourly, and bundles a 12-month subscription. We don't pitch this — if you want it, email us.
Peekr Cloud is free up to 10k spans/month, no card required. Paid tiers start at $29/mo (Starter) to $399/mo (Scale) by span volume. Extremis Cloud pricing is usage-based on memory and recall volume — email us for the current band. Self-hosting both products is always free (MIT).
Contact us
Tell us which product you're interested in and what you're shipping — we'll get back within 24 hours. Or skip the form and book a demo directly.
Book a demo
30-min walkthrough · Cal.com
Pick a time →
hello@starkspherelabs.com
Send a note →
Open source
Extremis · Peekr
Read the code →
Response time
< 24h on weekdays
Async-friendly across timezones
Registered as StarkSphereLabs Fz LLC · UAE free-zone LLC · invoicing globally · USD / AED / GBP / EUR.