01 / 05Shipped
Structured company memory
A queryable layer of decisions, exceptions, people, projects, and policies. Answer 'what did we decide about X' in one query, not in three meetings.
Live in production on the multi-tenant runtimeTYPED MEMORY
Company memory for every AI agent
Give your AI agents the company knowledge they need to actually work.
Your agents stop starting from zero every day. They know your customers, your decisions, your processes, your team's institutional knowledge. They act on it.
Live multi-tenant runtimeOpen core · MIT substrate · commercial runtimeWorks with Claude, GPT, Gemini, Llama, or your own modelSingle connector built per pilotF100-ready security model
For engineers: read the full architecture.
What makes Mycelium different
The structural moat, in four lines.
Four properties no closed-SaaS memory layer ships together. Read them once. Each one collapses a class of vendor risk a F100 procurement committee already underwrites.
- 01Customer-owned
Plain-text memory
Every memory record is a markdown file you own. Read it in any editor. Grep it with the tools your team already runs. No proprietary on-disk format to reverse-engineer at renewal time.
- 02Bi-temporal
Git-versioned history
Every write produces a commit. Every decision carries a diff. Bi-temporal queries return the state of the memory layer on any past date. Audit-grade by construction.
- 03MIT substrate
Open-core architecture
MIT-licensed substrate at ai-brain-starter on GitHub. Commercial runtime adds multi-tenant auth, audit log, real-time event streams, webhook reliability. Same shape as Vercel and Supabase.
- 04Portable
Customer-owned deployment
Multi-tenant managed, single-tenant managed, or on-premise install. Same runtime, different boundary. No exit strategy needed because the memory layer never left your perimeter to begin with.
These four lines are the answer when a F100 CIO asks what we ship that the bundled memory in their agent runtime cannot. The structural answer is on this page; the controls evidence is on the Trust Center.
Build versus buy
Why not have your team build this?
Three answers a Fortune 100 CTO who's already evaluating both paths recognizes on first read. None of this is novel architecture. The cost is in the boring parts that compound across years.
- 0112-18 months
Engineering time, on a problem that isn't your business.
A neutral, multi-tenant, audit-grade memory layer is a 12 to 18 month senior-engineering build: auth, tenant isolation, audit retention, write-back discipline, retry queues, vendor-specific signature verification, connector hardening, schema evolution. Your team has the skill. The opportunity cost is the customer-facing roadmap that ships instead.
- 02Permanent
The maintenance tail is permanent.
Connectors break as vendor APIs drift. Audit-log retention shifts under SOX, HIPAA, GDPR. Graph migrations come around every two years. RBAC reviews compound as headcount grows. Once you own the layer, you own all of it forever, on top of whatever else the team is building.
- 03MIT substrate
Open core inverts the build-versus-buy.
The substrate is MIT at github.com/adelaidasofia/ai-brain-starter. If you want your team to build the runtime around it, you can; nothing about the substrate is gated. If you want us to operate it for you, the commercial runtime ships with the connectors, audit, multi-tenant auth, and reliability already hardened. The decision is operate-or-not, not build-or-not.
- 04Neutral
Single-vendor bundles aren't the same alternative.
Waiting for OpenAI, Anthropic, or Google to ship a memory layer ties yours to one vendor's data residency, model release schedule, identity stack, and price card. The portability your procurement team asks about (data exportable, in plain text, on demand) does not survive that bundling. Mycelium stays neutral across model vendors by construction.
The math, at the margin: a 12-month internal build at $200K loaded per engineer across two engineers lands above the Enterprise tier in year one and stays there forever, with the maintenance tail counted as zero. The substrate is free either way.
Why this matters
Your AI agents are smart, but blind.
They don't know your customers, your decisions, your processes, or your team's institutional knowledge. So they hallucinate, redo old work, and embarrass you in front of clients.
The two paths most enterprise teams reach for first don't close the gap. Building it internally takes a year-plus of senior engineering time on a problem that isn't your business. Waiting for OpenAI or Anthropic to ship a memory layer bundles your data into one vendor's residency, release schedule, and identity stack.
Mycelium is the memory layer underneath: the platform your agents read from, write back to, and get smarter on, across every team and every tool you already use.
Memory-layer specialists
Mem0, Zep, Cognee, MemoryLake, Honcho, Letta, Memco, Graphlit
Each ships parts of the layer. None ship plain-text customer-owned, Git-versioned, bi-temporal, multi-tenant, audit-logged, open-core, multi-vendor model-portable memory together.
Retrieval and chat-on-docs
Pinecone, Weaviate, Qdrant, Glean, Notion AI, Coda AI
Search a corpus that already exists. Mycelium is upstream: the layer that creates the corpus, types it, version-controls it, and scopes it per role.
Agent runtimes with embedded memory
LangChain with LangMem, LlamaIndex, CrewAI, Cognition Devin
Memory inside an orchestration stack, not a neutral substrate across heterogeneous agents and model vendors.
Big-cloud incumbents
Vertex AI Memory Bank, Bedrock Knowledge Bases, Copilot Studio, Snowflake Cortex Agents, Salesforce Agentforce
Each runs inside a single cloud and a single identity stack. Mycelium is the portable substrate underneath.
Why teams buy Mycelium
What changes the week after install.
Procurement signs the engagement when the line items map to numbers their CFO already tracks. Below is the shape of what changes; specific magnitudes are scoped per pilot against your baseline metrics in Week 0.
Pain
AI agents start every conversation from zero.
Cost today
Sales reps re-discover account history every quarter. Support agents apologize for context that was already shared. Engineers solve problems another team solved last quarter.
After Mycelium
Agents act on Day 1 with the same context your senior people carry. The memory layer survives session boundaries, model swaps, and team turnover.
Pain
New-hire ramp takes three to six months.
Cost today
Each new hire's first quarter is a sunk cost. Multiply by your annual hiring volume.
After Mycelium
New hires read the memory layer their team already populated and ramp in week one. The memory layer survives session boundaries, model swaps, and team turnover. Specific ramp-time deltas are scoped per pilot against your baseline metrics.
Pain
Decisions get archaeological after a quarter.
Cost today
What did we decide about the SLA carve-out? Who approved the discount? Three meetings to reconstruct one decision. Compounds across legal, finance, sales, and operations.
After Mycelium
Every decision lands in a typed memory entry with the actor, the date, the rationale, and the artifacts it touched. One query returns the answer. Auditors get a binder; operators get an answer in seconds.
Pain
Compliance binders take weeks to assemble.
Cost today
SOC 2, vendor risk responses, audit observations: weeks of cross-team document chase per response. Bottlenecks the close of every Q4 enterprise deal.
After Mycelium
An agent reads HR records, IT tickets, and policy documents from the connected systems and produces an auditor-ready binder. Time-travel queries answer 'what was true on the audit observation date' without reconstructing anything by hand.
Pain
Your AI investment is locked to one vendor's memory layer.
Cost today
Switching cost grows monthly. Procurement notices on contract renewal.
After Mycelium
Memory is plain-text under your control. Swap Claude for GPT for Gemini for Llama or your own model without rebuilding the layer underneath. Vendor independence is a security control.
Pain
Knowledge silos mean teams duplicate work.
Cost today
Two product teams discover the same customer pattern. Two compliance reviews ask Legal the same question. Sales offers a discount Finance just rejected for another account.
After Mycelium
The memory layer is a single source across teams, scoped per role. HR writes; Engineering reads what's relevant. Salesforce notes feed the agent that drafts outbound. Audit log records who saw what, when.
Specific ROI numbers are modeled in Week 0 of the engagement against your existing metrics: rep ramp time, decision-cycle time, compliance binder assembly time, agent error rate, switching cost. We do not publish other customers' numbers without permission and we do not invent them.
How it works
Five components your team can actually explain.
Each component has a plain-language job. The technical detail underneath is for the engineer on your buying committee.
Memory · Routing · Skills · Learning · Portability
- 01
Memory store
Your enterprise's context lives in plain text files you own. Auditable. Version-controlled. Readable by a human or a model.
EngineerMarkdown over Git, typed-memory schema, no proprietary database lock-in.
MEMORY
- 02
Smart routing
Agents find the right context without scanning everything. Answers can be scoped to what was true on a given date.
EngineerRESOLVER pattern with bi-temporal lookup over the typed graph.
ROUTING
- 03
Skills
Reusable instructions your agents follow. Composable, testable, swappable. Not a prompt library.
EngineerEach skill ships with a single SKILL.md spec, an entry contract, and a reproducible side effect.
EXECUTION
- 04
The corpus gets richer
Every refined answer joins the brain. Next week's queries land on a deeper substrate, not a smarter model.
EngineerClosed-loop write-back discipline. Episodic memory promotes to procedural in seconds.
FEEDBACK
- 05
Model freedom
Swap Claude for GPT for Gemini for Llama for whatever's next. Your memory stays put.
EngineerSeparation of context from compute. Memory is portable; runtime is configurable per workflow.
PORTABILITY
Connectors
Plug into the systems your team already lives in.
Connectors normalize external context into the memory layer. The roster covers the systems your operators already work in every day.
16 public MCP servers shipped
New connectors land weekly
May 2026 wave: slack, imessage, parse, luma, graph-query, github plus ten prior. Each one is a public MCP package on the registry, installable in one command, hardened against the same write-back discipline as the production runtime.
Live in production
Communication·6 live
Slack
Messaging
Microsoft 365
Email + Docs + Teams
Google Workspace
Email + Docs + Drive
Gmail
Email
WhatsApp
Messaging
Zoom
Meetings + transcripts
Knowledge & Docs·4 live
Notion
Docs + wiki
Confluence
Wiki
Box
Files
Dropbox
Files
Engineering·4 live
GitHub
Code + issues
GitLab
Code + issues + pipelines
Linear
Issues + projects
Jira
Issues + projects
CRM & Sales·2 live
Salesforce
CRM
HubSpot
CRM + marketing
Business systems·5 live
Workday
HR + Finance
ServiceNow
ITSM
SAP
ERP
Oracle
HCM + ERP + Finance
ADP
HR + Payroll
Data & Analytics·2 live
Snowflake
Data warehouse
Databricks
Data + ML
Custom builds
Need a system that isn't here?
Name it during pilot scope. Custom connectors land in the same multi-tenant runtime, with the same vendor-specific signature verification, the same audit log, and the same retry plus dead-letter queue as the twenty-three above.
Scope a custom connectorEvery connector ships with vendor-specific signature verification (HMAC-SHA256, JWT, bearer tokens, CloudEvents), async retry queue with exponential backoff, dead-letter folder per tenant, and admin-only replay.
What ships
Five things your team gets out of the box.
Every line below is shipped, in production, today. Click through to /technical for the engineering detail.
02 / 05Shipped
Connect everywhere your team works
Twenty-three connectors live in production across Communication, Knowledge, Engineering, CRM, Business systems, and Data: Slack, Microsoft 365, Google Workspace, Gmail, WhatsApp, Zoom, Notion, Confluence, Box, Dropbox, GitHub, GitLab, Linear, Jira, Salesforce, HubSpot, Workday, ServiceNow, SAP, Oracle, ADP, Snowflake, Databricks. Anything else gets built during pilot scope.
- Slack
- Microsoft 365
- Google Workspace
- Gmail
- Zoom
- Notion
- Confluence
- Box
- Dropbox
- GitHub
- GitLab
- Linear
- Jira
- Salesforce
- HubSpot
- Workday
- ServiceNow
- SAP
- Oracle
- ADP
- Snowflake
- Databricks
Twenty-three live connectors with a custom-build path for what isn't hereINGESTION
03 / 05Shipped
Auto-summarize threads, PRs, meetings, decisions
A merged pull request becomes a workflow your team can reuse. A resolved Slack thread becomes a memory entry your agents can look up. No manual SOP-writing.
Synthesizers verified on golden-pair eval fixtures, 97.8 / 100 averageSYNTHESIS
04 / 05Shipped
Answer questions about any point in time
What did the team know on the day the audit observation happened? What did the policy actually say last quarter? Both questions, one query.
Bi-temporal resolver with public specTIME TRAVEL
05 / 05Shipped
Gets smarter every day, automatically
Every refined answer, every approved decision, every resolved exception writes back into the memory layer. The system improves without retraining a model.
Closed-loop daemon shipped, scheduled fallback in placeLEARNING
Production grade
What your security team actually wants to see.
Plain-language guarantee on top, the technical detail in mono underneath, for the security engineer reading along.
- 01
If you're Citibank, your competitor can't see your data. Ever. Different locks, different keys, separate spaces.
EngineerMulti-tenant JWT auth scoped per tenant. No cross-tenant reads, no shared state.
- 02
Compliance can answer who-did-what-when.
EngineerFull audit log: actor, tenant, timestamp, source on every read, write, and synthesis.
- 03
Your other tools learn the moment something happens, not at the end of the day.
EngineerPer-tenant Server-Sent Events. Subscribe to ingestion, synthesis, and decision events.
- 04
Hackers can't fake a message pretending to be Salesforce or Slack.
EngineerHMAC-validated webhooks. Async retry queue with exponential backoff. Dead-letter folder per tenant. Admin-only replay.
- 05
Reliability you can verify in writing.
EngineerEvery release passes 59 automated checks before it ships. End-to-end integration test on every code change. Our reliability standards and the step-by-step recovery procedure are public on GitHub.
59 automated checks per release · End-to-end integration tested on every change · 23 live connectors in production
Security & Compliance
What procurement asks first.
Send this section to your security team. Every line maps to a control they expect to see.
- 01available
Data residency
US and EU options on Vercel + AWS-backed deployments. Region pinning configurable per tenant on the productized runtime.
- 02available
Deployment options
Multi-tenant managed runtime, single-tenant managed runtime, and on-premise install. Single-tenant and on-premise unlock with the Mycelium Enterprise tier.
- 03enterprise
Service-level agreement
Enterprise tier includes a contractual SLA. Uptime target and response-time tiers are negotiated per pilot scope and named in the engagement letter.
- 04available
Role-based access control
Permissions scoped per tenant, per workspace, per resource. Read, write, and synthesis are independent. Audit log carries the actor identifier on every event.
- 05available
Audit log retention
Default 12 months in the managed runtime, extendable to 7 years for SOX or HIPAA-adjacent buyers. Audit exports are gzipped JSONL signed with a tenant key.
- 06available
Encryption
TLS 1.3 in transit. AES-256 at rest. Per-tenant encryption keys available on the Mycelium Enterprise tier; key rotation via the admin API.
- 07available
Vulnerability disclosure
Reach security@mycelium-ai.co. We acknowledge within one business day, triage within five, and credit researchers in the public CHANGELOG.
What you build on it
Three workflows your team can run on Monday.
These are the kinds of workflows the memory layer makes routine. Each one depends on memory that survives session boundaries, tools, and time.
- 01
Sales outbound that knows the account
An agent drafts your sales outbound knowing every Salesforce note about the prospect, every conversation your team has had with their company in the last year, and the last successful close in the same segment. Your rep edits one sentence and sends.
- Salesforce
- Slack
- Notion
- Gmail
- Memory layer
- Agent → Draft
- 02
SOC 2 audit binder, prepared
An agent prepares your SOC 2 audit binder by pulling HR records, IT tickets, and compliance policies into a single document your auditors approve in one pass. It answers what was true on the audit observation date, even if the policy has since changed.
- Workday
- ServiceNow
- Confluence
- Memory layer
- Agent → Binder
- 03
New-hire onboarding in week one, not month three
An agent onboards a new hire by surfacing every decision, every relevant Slack thread, and every meeting note from the team they're joining. They walk in on day one knowing what was already tried and why.
- Slack
- Notion
- Linear
- Gmail
- Memory layer
- Agent → Briefing
Where Mycelium sits
Memory layer, not a search box.
The systems below solve real problems. None solve the memory-layer problem at the level a Fortune 100 buyer can put on procurement. Columns are what an enterprise buyer actually evaluates; rows are scored against each vendor's public documentation as of 2026-05.
| Your data stays portable | Tracked in version control | Multi-tenant safe | Audit-logged | Swap models without rebuilding | Open core | Enterprise connectors | Answers are version-aware | |
|---|---|---|---|---|---|---|---|---|
| Mycelium | ||||||||
| Mem0 | · | · | · | |||||
| Zep | · | · | · | |||||
| MemoryLake | · | · | · | |||||
| Cognee | · | · | · | · | ||||
| Memco | · | · | · | · | ||||
| Graphlit | · | · | · | · | ||||
| Weaviate | · | · | · | · | ||||
| Hyperspell | · | · | · | · | · | |||
| Honcho | · | · | · | · | · | |||
| Letta | · | · | · | · | · | · | ||
| Pinecone | · | · | · | · | · | |||
| Glean | · | · | · | · | · | |||
| Notion AI | · | · | · | · | · | · |
Verified against each vendor's public documentation as of 2026-05. Corrections to contact@mycelium-ai.co.
Proof
In production today.
Live deployments on the multi-tenant runtime. Twenty-three connectors active. Every release passes 59 automated checks before it ships.
23live connectors. Slack, Microsoft 365, Google Workspace, Gmail, WhatsApp, Zoom, Notion, Confluence, Box, Dropbox, GitHub, GitLab, Linear, Jira, Salesforce, HubSpot, Workday, ServiceNow, SAP, Oracle, ADP, Snowflake, Databricks. Custom builds available during pilot scope.
236 / 236automated tests passing. End-to-end integration test runs on every code change. The build does not ship if any check fails.
59checks per release. Reliability gates run before every release. Our reliability standards and step-by-step recovery procedure are public on GitHub for your security team to review.
MITopen core. Memory layer ships open-source. Productized runtime ships private. Same shape as Vercel and Supabase.
Tenant 0is us. Mycelium runs Mycelium. Our team's CRM, weekly digest, decision log, and connector queue live in the same runtime we sell. We see the bug before the customer does.
Co-Founders
Built by operators, not researchers.
Adelaida Diaz-Roa
Co-Founder
Seven companies. Two exits. Shark Tank at 18 (Mark Cuban + Barbara Corcoran). Success Magazine 30 Under 30.
Authored ai-brain-starter (MIT, in production), the open-source substrate Mycelium productizes, and has been running it across her own active companies since before it had a name. Live install today: 15,655 typed memory entries, 51,236 relationships, 389 CRM contacts with full interaction history, 129 decisions logged with provenance.
Leads product, architecture, install delivery, and the open-source substrate.
Nelly Ortiz
Co-Founder
A decade-plus at the enterprise tier. Director of Business Hacking US Co-Lead at Globant, Head of Digital Strategy USA at NTT DATA, IBM iX Senior Strategy Consultant. Advised IBM, Citi, Disney, MGM, L'Oréal, P&G, Univision, Intuit, NTT DATA, Globant, Mindtree, and dozens of Americas-wide enterprises on digital and AI strategy.
TEDx Santa Barbara speaker on closing the LATAM wealth gap with technology. Globant Awards Tech Entrepreneur. Leads enterprise sales, partnerships, cohort delivery, and brand.
Pricing
Three tiers. One memory layer.
Mycelium Personal is a monthly subscription on the shared Mycelium WhatsApp line; your phone number is your identity. Mycelium Lite and Enterprise are annual subscriptions with a dedicated WhatsApp Business number per tenant, plus the full connector surface. Platform license invoiced annually, usage billed monthly against active connectors and query volume. Multi-year prepay reduces total cost: three-year prepay 15% off, five-year prepay 20% off plus early access to vertical packs at general availability. Specific usage unit rates are named in the engagement letter.
Second brain in WhatsApp
Mycelium Personal
$97
/ month
Annual: $970, two months free. Founding cohort: $49 / month locked-in for life, first 100 subscribers.
Same-day activation after WhatsApp number registration
Operators, writers, founders, and solo consultants who carry context across investors, board, contractors, family, and creative work and want it queryable from WhatsApp.
- Shared Mycelium WhatsApp line; your phone number is your tenant identity
- Your vault accessible from WhatsApp text and voice notes
- Multi-model routing: Claude default, GPT, Gemini, and Llama via per-turn override
- Five active skills (journal, weekly insights, meeting notes, decision log, follow-ups)
- 30-day rolling memory window with per-subscriber encryption
- Identifies itself as an AI assistant on first contact, every new conversation
- ai-brain-starter substrate compatibility; your vault stays portable
Month-to-month, cancel any time. Founding price locks for the lifetime of the subscription.
Join the founding cohortAI-Native Sprint
Mycelium Lite
$25K
/ year platform + usage
Or $2,500 / month, 12-month minimum
60-day install
Scale-ups with an AI team adopting agents internally and a small connector surface.
- 5-day install + 60 days of office hours (no dedicated engineer)
- Up to 3 connectors active at install (no custom builds)
- Multi-tenant managed runtime (shared)
- Email support, response within 1 business day
- Best-effort uptime, no contractual SLA
- Standard handoff through your security team
- Public reliability standards and step-by-step recovery procedure
12-month minimum term. Multi-year prepay available.
Start a Lite pilotProcurement-grade
Mycelium Enterprise
$150K – $500K
/ year platform + usage
90-day install + 12-month runtime
Fortune 100 buyers with an existing AI initiative, a real Vendor Risk Management process, and a named integration scope.
- Dedicated install engineer assigned for the full 90 days
- Up to 15 connectors active, plus custom connector builds scoped on Week 0
- Single-tenant managed runtime, or on-premise deployment in your cloud
- Dedicated Slack or Teams support channel, response within 4 business hours
- Contractual SLA with named uptime targets + response-time tiers
- White-glove handoff through your Compliance, Security, and Legal teams
- Quarterly business review with leadership
- Per-tenant encryption keys + key rotation via admin API
- Same public reliability standards and step-by-step recovery procedure
12-month minimum term. Multi-year prepay available with vertical-pack early access at GA.
Talk to salesMycelium Vertical (Legal, Finance, Healthcare) ships Q3 2026 as a procurement-grade extension of the Enterprise tier. Multi-year prepay buyers get early access at general availability. Join the waitlist on the inquiry form.
How a pilot runs
From scoping call to production handoff.
Every engagement follows the same shape. Mycelium Lite compresses the install to 60 days, then 12 months of runtime under the annual platform license. Mycelium Enterprise runs the full 90-day install with a dedicated engineer, then 12 months of runtime. Both tiers renew annually with usage billed monthly.
Week 0
Scoping
Scoping call, connector audit, security review with your team. We name the install scope, the success metric, and the named go-live date. Engagement letter signed before Week 1 starts.
Week 1 – 2
Install + first connectors
Memory layer installed in your tenant. First three connectors activated against your real data. Initial schema seeded from your domain.
Week 3 – 4
Schema customization
Memory schema customized for your decision flows, exceptions, and terminology. Routing calibrated against the queries your operators actually ask.
Week 5 – 6
Agent integration + closed-loop training
Your agents read from and write back to the memory layer. The system promotes refined answers into procedural memory automatically. First production traffic.
Week 7 – 8
Production handoff
Production handoff with a 60-day retrospective on day 90. Reliability standards reviewed with your team, step-by-step recovery procedure signed off, Compliance handoff complete.
Frequently asked
Procurement asks. We answer.
Eleven questions an enterprise buyer asks before booking a call. If yours isn't here, write contact@mycelium-ai.co.
What does Mycelium actually ship?
A multi-tenant memory-layer runtime, twenty-three live connectors across Communication, Knowledge, Engineering, CRM, Business systems, and Data, a typed-memory schema, a bi-temporal RESOLVER, an HMAC-validated webhook receiver layer with retry queue and dead-letter routing, an append-only audit log with per-tenant scoping, and a Server-Sent Events stream. Open core under MIT for the substrate; commercial license for the productized runtime.
If the open-source repo is free, what does Mycelium actually sell?
The free repo (ai-brain-starter, MIT) is the recipe. It teaches the pattern, you can clone it and run it on a laptop. Mycelium sells what the recipe doesn't include. The hosted runtime that runs 24/7 with HMAC-validated webhook receivers, tenant-scoped JWT auth, audit log, and dead-letter routing. That's the production-grade infrastructure no buyer wants to operate themselves. The install: connector configuration, knowledge-graph densification, integration with their stack, training the team. The custom connectors for systems the open core doesn't ship with. Same shape as Vercel and Supabase: Next.js and Postgres are open source, companies pay Vercel and Supabase to host, scale, and operate them. The repo is the recipe. Mycelium is the chef, the kitchen, the security cameras, and the guarantee it doesn't burn while you're in another meeting.
What does it cost?
Mycelium Personal is $97 per month or $970 per year. Founding cohort is $49 per month locked-in for life, capped at the first 100 subscribers. Mycelium Lite is $25K per year for the platform license, plus usage billed monthly against active connectors and query volume. Mycelium Enterprise is $150K-$500K per year for the platform license, plus usage. Lite and Enterprise carry a 12-month minimum term. Three-year prepay is 15% off; five-year prepay is 20% off plus early access to vertical packs at general availability. Specific usage unit rates and per-customer SLA terms are named in the engagement letter, after a scoping call.
How long does install take?
Mycelium Lite is a 5-day install plus 60 days of office hours, then runtime under the annual platform license. Mycelium Enterprise is 90 days with a dedicated install engineer assigned the full window, then 12 months of runtime. Both end with a production handoff and a 60-day retrospective.
Can it run inside our cloud or on-premise?
Mycelium Enterprise yes: single-tenant managed runtime in your AWS, GCP, or Azure account, or fully on-premise inside air-gapped infrastructure. Mycelium Lite is multi-tenant managed only.
What's your SOC 2 / HIPAA / GDPR posture?
GDPR: yes, the runtime is built to be processor-grade with the controls in our DPA. HIPAA: BAAs are negotiable on the Mycelium Enterprise tier; not in scope on Mycelium Lite. SOC 2 Type II: not certified yet. Readiness program is planned for 2026 with an audit window targeted for 2027. We don't claim what we don't have.
Does it work with our existing AI stack?
Yes. Mycelium is model-agnostic. Memory lives as plain-text typed entries you own; agents under Claude, GPT, Gemini, Llama, or your own model read from the same layer. Vector stores like Pinecone, search tools like Glean, and orchestration platforms like LangChain are upstream or downstream of Mycelium, not replaced by it.
What if a connector we need doesn't exist?
Custom connectors are scoped on Week 0 of a Mycelium Enterprise engagement. The connector becomes a skill in your tenant; the source is shared as part of the deliverable so you can audit, fork, or maintain it after handoff.
Who owns the data? Can we export everything?
You own it. Tenant data lives as plain-text Markdown under your vault root, version-controllable in your own git remote if you choose. There is no proprietary database lock-in. Audit-log exports are available on demand as gzipped JSONL signed with a tenant key.
What happens to our install if Mycelium goes away?
Open-core architecture is our answer to vendor risk. The substrate (schemas, RESOLVER specification, skill contract, eval framework, reference connector implementations) is open-source under MIT at github.com/adelaidasofia/ai-brain-starter. Your tenant data, schema, and skills survive in plain text. The productized runtime is replaceable with a self-hosted equivalent built on the open core. Same shape as the Vercel + Next.js relationship.
How is this different from mem0, Letta, Glean, Pinecone?
Different category for each. mem0 / Letta / LangMem / Zep tune chatbot memory at the conversation layer. Pinecone / Weaviate / RAG frameworks search a corpus that already exists. Glean / Notion AI search what's already written. Mycelium is the typed, version-controlled, write-back memory layer those systems read from and write back to. The Comparison section earlier on this page maps the axes.
Every answer above is what we actually ship today. When something changes, this section changes with it.
Open core
Free core. Private runtime.
The memory layer started as a working prototype before it became a commercial runtime. Months of production use across a working team hardened the architecture into the open-source substrate that ships today.
That substrate is ai-brain-starter at github.com/adelaidasofia/ai-brain-starter, MIT-licensed, in production. Free to fork, free to extend, free to install on a single operator's machine in two hours.
The productized runtime that operates that memory layer across an enterprise, with multi-tenant auth, audit, real-time event streams, and webhook reliability, ships under a commercial license. Same shape as Vercel and Supabase: free core teaches the pattern, paid runtime carries it across teams.
Mycelium runs Mycelium. Our team's CRM, weekly digest, decision log, and connector queue live in the same runtime we sell. We see the bug before the customer does.
adelaidasofia / ai-brain-starter
MITThe open-source core, in production. Drop into a folder, run one command, install on your machine in about two hours of operator time end-to-end.
- 11-step E2E integration test green on every push
- 6 ingestion connectors with shared utilities
- 2 autonomous synthesizers (PR · Slack)
- Closed-loop learning daemon
- Reliability Manifesto + Recovery runbook public
Pilot
If your team is adopting AI agents, talk to us.
Rolling enrollment for Mycelium Lite and Mycelium Enterprise. Mycelium Vertical opens for general availability in Q3 2026; pick the waitlist option below to reserve a slot. Sixty to ninety days from scoping call to a working memory layer in your tenant.
Mycelium Lite
$25K / year platform + usage · 60-day install
Mycelium Enterprise
$150K – $500K / year platform + usage · 90-day install + 12-month runtime
Mycelium Vertical
Available Q3 2026 · Legal, Finance, Healthcare · waitlist on the inquiry form
Tenant model
Multi-tenant JWT · audit · SSE · retry · DLQ
Connectors at install
Up to three on Lite; up to fifteen on Enterprise plus custom builds
Multi-year prepay
3-year 15% off · 5-year 20% off plus early access to vertical packs
Cadence
Rolling enrollment · cohort waves quarterly
Or write us directly at contact@mycelium-ai.co. Security teams: security@mycelium-ai.co.