Saprolings · AI Studio + Web3 Studio

We build and ship AI systems and on-chain products.

Saprolings is a development and infrastructure partner with two studios. AI Studio builds enterprise brains, and AI agents that work alongside your team like real people. Web3 Studio takes tokens, on-chain apps and raises from first design to launch. Engineers run every engagement.

Talk to us → Book a call See the studios
60+
Web3 teams advised
$350M+
Combined market cap of the projects we've worked with
50+
Exchange relationships
3,300+
Notes in the enterprise brain we run in production
7,000+
Wikilinks joining that knowledge into one graph
87
Agent skills in production
01 / Studios

Two studios, one engineering bench.

We don't hand over a strategy deck and leave. Every engagement ends with something running: a system in your stack, a contract on-chain, a data room in investors' hands.

2 services

AI Studio

Company memory and AI agents that work alongside your team like real people, built into your stack.

  1. 01
    Enterprise brainA governed, linked company memory that agents load before they act
  2. 02
    Agentic workflows and systemsAI teammates, shared agent workspaces and the data infrastructure they run on
Inside AI Studio →
3 services

Web3 Studio

Token, product and capital work for on-chain teams, delivered by engineers.

  1. 01
    Token launch and advisoryToken design, launch planning and price-support forensics
  2. 02
    Blockchain app deploymentContracts, dApps, indexers and admin consoles, audited and operated
  3. 03
    FundraisingData rooms, due-diligence packs and investor targeting
Inside Web3 Studio →
02 / AI Studio

An enterprise brain your people and your agents both work from.

Most company AI starts every conversation from zero. An enterprise brain gives it a memory it can check: governed, linked and sourced. We run one in production ourselves, and it sits behind every agent we deploy.

Service 01

AI enterprise brain

Nine capabilities, all live in the brain we run on ourselves.

Exhibit 1How an enterprise brain turns daily work into memory agents can use
Sources flow through a four-stage ingest pipeline into a governed vault, then a memory layer, then agents and people; sessions write back new notes and skills. 01 · SOURCES Your team's work Tickets Meetings Team chat Inbox captures Documents 02 · INGEST Four-stage pipeline Reduce Reflect Reweave Verify → task tracker 03 · VAULT Governed brain Provenance on every note Write tiers at commit Recursive summaries Wikilink graph Full git history 04 · MEMORY Memory layer Sideloaded working memory Semantic index Session recall Four retrieval engines 05 · AGENTS Agents and people 87 skills 11 specialist agents Your team's AI tools Answers with sources Continuous learning: every session writes back new notes and skills Reconciled by date and authority · confidence decays unless re-verified · every change through a reviewed pull request
  1. 01 · SourcesYour team's workTickets · Meetings · Team chat · Inbox captures · Documents
  2. 02 · IngestFour-stage pipelineReduce · Reflect · Reweave · Verify · → task tracker
  3. 03 · VaultGoverned brainProvenance on every note · Write tiers at commit · Recursive summaries · Wikilink graph · Full git history
  4. 04 · MemoryMemory layerSideloaded working memory · Semantic index · Session recall · Four retrieval engines
  5. 05 · AgentsAgents and people87 skills · 11 specialist agents · Your team's AI tools · Answers with sources

Continuous learning: every session writes back new notes and skills.

Reconciled by date and authority; confidence decays unless re-verified; every change through a reviewed pull request.

01

Memory sideloading

Agents start every session already briefed. Pinned working memory, a semantic memory index and the project's own files load in a fixed order before the first prompt, so nobody pastes the background in by hand.

Context in place before the first prompt
02

Wikilinked knowledge graph

Each note holds one idea and links to the ideas it depends on. Link-integrity checks and duplicate-name resolution keep the graph clean, and maps of content give every domain a front door.

3,300+ notes · 7,000+ wikilinks
03

Recursive summaries

Every folder carries a summary of everything beneath it, so an agent reads down the tree instead of reading every file. Adding a thousand notes barely changes what an agent has to read.

590+ summaries · O(log n) navigation
04

Four retrieval engines

Hybrid semantic and keyword search, semantic memory recall, full-text search over past agent sessions, and exact match. Agents pick the engine that fits the question.

~6,000 indexed passages
05

Provenance on every note

Every note is tagged human-written, AI-generated or AI-modified. Agents treat any number without a source as a hallucination.

Provenance tags on every note
06

Governed writes

Folders are tiered human-only, human-primary or open, and the tiers are checked on every commit. Every change lands through a reviewed pull request, with full history for every note.

450+ commits · 11 contributors · PR-only
07

Reconciliation and decay

When two notes disagree, the newer sourced note wins, and on a tie the human-authored one does. Agents flag conflicts instead of guessing. Knowledge nobody re-verifies loses confidence over time.

High → medium at 90 days → low at 180
08

Always-on ingestion

Tickets, meetings, team chat and inbox captures run through a four-stage pipeline (reduce, reflect, reweave, verify) and push actions to your task tracker. A glossary check keeps everyone using the same words.

630+ ticket notes · 330+ meetings · 430+ captures
09

Continuous learning

Each working session ends up as atomic notes and reusable skills. Gap scans surface terms the team uses but hasn't written down, and short question drips fill the thin spots.

87 skills · 11 specialist agents
What you get
  • A governed knowledge vault in your own git, with provenance and write tiers
  • A command-line toolkit for search, validation, summaries and gap scans
  • A library of agent skills and specialist agents, tuned to your domain
  • Ingestion connectors for your tickets, meetings, chat and inbox
  • A memory layer wired into the AI tools your team already uses
  • Runbooks and hands-on training, so your team owns it from day one
Service 02

Agentic workflows and systems

AI agents that work alongside your team like real people: they take tasks, share channels, hand work off and report what they did. We deploy the workspace they share, and the data they run on.

Exhibit 2A shared agent workspace: agents act, but never hold a key
People and agents work in shared channels; actions go through an execution broker that only accepts signed, whitelisted agents; agents read a read-only data lake with one-hour credentials. 01 · PEOPLE + AGENTS Side by side Desktop app on every machine Keys stay on the device Bring your own model Plain English is enough 02 · WORKSPACE Shared channels People and agents together Task queue and history One community per client Daily reports from agents 03 · BROKER Execution broker Signed, whitelisted agents Rules enforced server-side Agents never see a key Activity monitored 04 · ACTIONS What gets done Exchange and on-chain orders Research and backtests Reports and dashboards Tickets and hand-offs 05 · DATA LAKE Read-only · raw → clean → research-ready · nightly refresh one-hour credentials
  1. 01 · People + agentsSide by sideDesktop app on every machine · Keys stay on the device · Bring your own model · Plain English is enough
  2. 02 · WorkspaceShared channelsPeople and agents together · Task queue and history · One community per client · Daily reports from agents
  3. 03 · BrokerExecution brokerSigned, whitelisted agents · Rules enforced server-side · Agents never see a key · Activity monitored
  4. 04 · ActionsWhat gets doneExchange and on-chain orders · Research and backtests · Reports and dashboards · Tickets and hand-offs
  5. 05 · Data lakeRead-only dataRaw → clean → research-ready · Nightly refresh · One-hour credentials for agents
02.1

AI teammates

Agents that work alongside your team like real people.

  • Each agent has a name, a role and a seat in your team's channels
  • They pick up tasks, hand work off and ask a human when a call is theirs to make
  • Every agent posts what it did to a shared channel, every day
  • Meetings, tickets and chat turned into notes, decisions and tasks without anyone typing them up
11 specialist agents already work alongside our own team
02.2

Shared agent workspaces

People and agents in the same channels, and no agent ever holds a key.

  • A desktop app for every teammate, joined through a shared workspace of channels, queues and history
  • Agent runs execute on each person's own machine, so keys and cached data never sit on a server
  • Actions go through an execution broker that accepts only signed, whitelisted agents and enforces the rules
  • Bring your own keys, model, compute and agent tooling, open-weights models included
Plain English is enough: non-coders direct agents to research, test and act
02.3

Data infrastructure for agents

A data lake your agents can query, but never corrupt.

  • Multi-source ingestion, layered from raw to clean to research-ready
  • A data-quality ledger and a nightly refresh
  • Read-only, scoped access for people, clients and agents, queried in place
  • Short-lived credentials for agents, so a leaked key expires within the hour
Agents can query the lake but cannot change it
02.4

Plain-English analytics

Ask your data a question and get the answer with its evidence.

  • Natural-language to SQL agents over an MCP server your AI tools already speak
  • Every answer returns the statistics behind it and the query that produced it
  • A house rulebook the agent must follow, and a permanent log of every trial it runs
  • Guardrails that hold in production: human-owned rules and automatic circuit breakers
Every answer is traceable to the query behind it
03 / Web3 Studio

Tokens, on-chain apps and raises, taken from design to launch.

Token design that holds up once it trades, production dApps with audits behind them, and a raise that's ready before the first investor call.

Service 01

Token launch and advisory

Token design that holds up once it trades.

  • Token utility and supply design, drawn from a sixteen-mechanism options framework
  • Price-support forensics: why buybacks miss, and what moves the index your token trades against
  • Market-structure reviews pulled live from exchange APIs
  • Launch planning, exchange strategy and listing readiness
60+ Web3 teams advised · $350M+ client market cap
Service 02

Blockchain app deployment

Production dApps, architected, integrated, audited and operated.

  • Vault and DeFi app stacks: smart contracts, web app, event indexer, admin console and backend
  • Independent security audit before mainnet
  • Deployment, listing on public TVL trackers, and day-to-day operation
  • Multisig custody and permissioned operations
On-chain vault live on Arbitrum since July 2026
Service 03

Fundraising

Raise-ready before the first investor call.

  • Investor data rooms: a thirteen-section structure with disclosed limitations and risk factors
  • Risk frameworks covering smart contract, key management, operational security and incident response
  • Due-diligence response packs, written and ready
  • Investor targeting, tiered outreach lists and exchange introductions
50+ exchange relationships
Exhibit 3How we take an on-chain app from design to live deposits
  1. 01ArchitectSystem design, threat model and delivery plan
  2. 02ContractsSmart contracts, tests and deployment scripts
  3. 03AuditIndependent security audit, findings closed before mainnet
  4. 04DeployMainnet deployment, multisig custody, web app live
  5. 05Index & listEvent indexer, admin console, TVL on public trackers
  6. 06OperateMonitoring, releases and day-to-day operation
04 / Work

Recent work, all of it in production.

Four engagements, two from each studio, with client names left out.

Web3 Studio · Blockchain app deployment

A DeFi yield vault, from architecture to live deposits

Problem
A protocol needed a production vault stack that could take public deposits, with investor-grade controls.
Shipped
We architected, integrated, audited and operated the full stack: contracts, web app, event indexer, admin console, multisig custody and the investor data room.
Result
Live on Arbitrum since July 2026, independently audited before mainnet, with TVL tracked on a public DeFi data aggregator.
Web3 Studio · Token advisory

Price-support forensics for an exchange-listed token

Problem
Two buyback programmes had spent seven figures, and neither held the price.
Shipped
We traced the perpetual's settlement index across its spot venues, pulled open interest live from exchange APIs, and mapped sixteen utility mechanisms that could replace the buybacks.
Result
The cause, in numbers: perp open interest above a quarter of circulating supply, and an index venue that a few thousand dollars could move by 5%.
AI Studio · Enterprise brain

A company memory for a digital-asset firm

Problem
Knowledge was scattered across tickets, meetings and chat, and every AI session started from zero.
Shipped
A governed brain in the firm's own git, with always-on ingestion, memory sideloading, provenance on every note and a library of skills and specialist agents.
Result
3,300+ notes and 7,000+ wikilinks, 630+ ticket notes and 330+ meetings captured, and every change reviewed through a pull request.
AI Studio · Agentic systems

An agent workspace for a strategy marketplace

Problem
Strategy curators who don't write code needed to research, test and run strategies without ever holding a trading key.
Shipped
A shared agent workspace driven in plain English, an execution broker that acts for whitelisted agents, and a read-only data lake behind short-lived credentials.
Result
English replaced code as the interface. Keys never leave the broker, and a leaked data credential expires within the hour.
05 / How we ship

One delivery model for both studios, from first call to production.

The same team takes your project from architecture to operation, so nothing gets lost between the people who design it and the people who run it.

  1. 01DiscoverWe map the problem, your stack and the result you need, in writing.
  2. 02ArchitectOne design covering data, contracts, agents and controls, agreed before we build.
  3. 03BuildSmall increments you can see working, reviewed through pull requests.
  4. 04ShipAudited, tested and deployed to production, with runbooks.
  5. 05OperateWe run it with you, or hand it over to a team we've trained.
Contact

Tell us what you need built.

Share the problem, the stack you have today and the date you need it live. You'll hear back from the team that builds it.

Email the team → Book a call