01 · Architecture
Taxonomy & folder design
A folder and tag structure mapped to how your business actually works (by function, client, process, and policy) so both staff and agents always know where a fact lives.
Foundation layer · 3–4 weeks · €4,500
Second Brain with Obsidian, the brain your AI agents read from.
An AI agent is only as good as the knowledge it can retrieve. We build your company a structured Obsidian vault, a plain-markdown knowledge base that every agent reads from and writes back to. One source of truth for your team and your AI workforce, owned by you, no vendor lock-in. See what an accurate knowledge layer is worth across a fleet of agents.
What is a Second Brain with Obsidian?
A Second Brain with Obsidian is a structured company knowledge base built as a vault of plain-markdown files that both your team and your AI agents read from and write back to. theagency47 designs the taxonomy, ingests and cleans your documents, links them so an agent can navigate by reference instead of guessing, and wires the vault into your agents as their shared, persistent memory. It is the foundation layer beneath an AI workforce, agents are only as accurate as the knowledge they retrieve, and the vault is that knowledge in a form both humans and machines can use.
theagency47 · Updated June 2026 The problem
Most AI projects fail at the knowledge layer, not the model.
The model is rarely the bottleneck. The bottleneck is that your company's knowledge lives in forty places, inboxes, Slack threads, a wiki nobody updates, three Google Drives, and people's heads. When an agent can't find the right answer, it does the worst possible thing: it invents one. A confident wrong answer from an agent costs more than no answer at all.
A knowledge base is the collection of documents an agent retrieves from at decision time, through retrieval-augmented generation (RAG). The quality of that knowledge base directly determines the quality of every agent's output. Build the brain first, and every agent you deploy afterward gets sharper for free.
What we build
A vault designed for humans and agents at the same time.
Not a dump of files in a folder. A navigable knowledge architecture with a deliberate taxonomy, dense internal links, and clean entry points an agent can reason over.
02 · Ingestion
Document ingestion & cleanup
We pull in your knowledge sources (policies, SOPs, FAQs, product specs, past correspondence) dedupe them, strip the noise, and rewrite them into clean, atomic notes. Volume is scoped with you in discovery.
03 · Linking
Wiki-style internal links
Notes are linked by reference, the way a researcher cross-references. An agent can follow links to gather context instead of embedding your entire corpus on every query.
04 · Agent wiring
MCP / retrieval connection
We connect your agents to the vault via the Model Context Protocol or a retrieval layer, with read paths and controlled write-back paths.
05 · Governance
Write-back & review rules
Clear rules for what agents may write, where, and what stays human-only, so the vault gets richer over time without drifting or filling with low-quality machine notes.
06 · Handover
Team enablement
A working session that teaches your team to maintain the vault, plus a maintenance playbook. The brain is only useful if it stays alive after we leave.
The loop
Agents read from it. Agents feed it. It compounds.
A static knowledge base goes stale the day it ships. A Second Brain is a living loop.
- Read: An agent searches the vault, pulls the relevant notes into its context window, and grounds its answer in those sources (citing them) rather than guessing.
- Escalate: If the relevant note is missing, a well-built agent does not invent an answer. It flags the gap, which tells you exactly where the brain needs to grow.
- Write back: Agents append meeting summaries, resolved tickets, decisions, and new facts under controlled paths, so the next task starts from more knowledge than the last.
- Single source of truth: Humans and agents edit the same markdown files. No human copy and separate AI copy quietly drifting apart.
This is why the brain comes first. Every agent in a Workforce Starter or Workforce Pro build plugs into the same vault and gets smarter as the vault grows.
Custom skills
Custom skills that turn the brain into action.
A vault is potential, skills are how agents put it to work. We build a starter set of custom Claude skills that operate directly on your vault, portable across your agents and your team's tools.
Skill · Retrieve
Knowledge concierge
Answers any staff or agent question strictly from the vault, with citations to the source notes, and escalates instead of guessing when the answer isn't there.
Skill · Capture
Intake & note-maker
Turns raw inputs (meeting transcripts, emails, tickets) into clean atomic notes, filed under the right path with the right links, so the brain grows without manual upkeep.
Skill · Maintain
Vault keeper
Runs hygiene on a schedule: flags stale notes, proposes merges for duplicates, and surfaces the gaps your agents keep hitting so you know where to grow the brain.
A starter set of skills is included in the build. More specialized skills, a weekly executive digest, a role-specific onboarding path, a compliance lookup, are built as add-ons or under a retainer. Every skill is yours to keep and runs on the open Model Context Protocol.
Why Obsidian
Plain markdown you own beats a database you rent.
We are tool-agnostic. We choose Obsidian for the agent knowledge layer for concrete reasons.
| Dimension | Obsidian (Second Brain) | Vector-DB RAG only | Notion / Confluence |
|---|---|---|---|
| Storage format | Plain markdown, owned by you | Embeddings in a hosted DB | Proprietary database |
| Vendor lock-in | None, it's just files | Tied to the DB / pipeline | Tied to the platform |
| Human-readable | Yes, staff use it too | No (vectors, not text) | Yes |
| Agent read + write | Native (same files) | Read-heavy; write is extra plumbing | API-mediated, rate-limited |
| Token efficiency (small–mid KB) | High (follow links, not dump corpus | Lower) chunk retrieval each query | N/A |
| Per-seat data cost | €0 (app is free) | Infra + ops monthly | Per-seat licensing |
| Best for | Agent memory + team wiki in one | Very large unstructured corpora | Human-only documentation |
For very large unstructured corpora, a vector layer still has its place, and we'll add one when the data justifies it. For the typical SMB knowledge base, a well-linked Obsidian vault is faster to build, cheaper to run, and something your team actually opens.
What's included
Everything to stand up the brain, and keep it alive.
- Knowledge discovery workshop (half-day)
- Custom vault taxonomy & tag system
- Ingestion & cleanup of your knowledge sources
- Deduplication, cleanup & atomic-note rewrite
- Wiki-style internal linking pass
- Agent wiring via MCP / retrieval layer
- Read paths + controlled write-back governance
- A starter set of custom Claude skills on your vault
- Maintenance playbook + team enablement session
- Full ownership, plain-markdown files, no lock-in
- Bundle with a Workforce build and save €1,500
Timeline
From scattered docs to a living brain in 3–4 weeks.
- Days 1–3: Knowledge discovery workshop + source inventory across your tools
- Days 4–7: Taxonomy & vault architecture design, signed off before ingestion
- Days 8–16: Ingestion, cleanup, atomic-note rewrite, and internal linking
- Days 17–21: Agent wiring (MCP / retrieval), read + write-back paths, governance rules
- Days 22–25: Retrieval testing against real questions, gap-filling, iteration
- Days 26–28: Team enablement session + maintenance playbook handover
Pricing & terms
One brain, one price, and a bundle saving when you build on it.
€4,500
Fixed foundation-layer build. Scope set in discovery. 3–4 weeks.
- Payment: 50% deposit at signing, 50% at handover
- Bundle saving: take €1,500 off the combined price when you add a Workforce Starter, Workforce Pro, or Enterprise build within 60 days, e.g. Starter + Second Brain = €10,500 instead of €12,000
- Ownership: Full transfer of the vault, plain-markdown files, no proprietary format
- Scope: We right-size the document set in discovery; unusually large or messy corpora are flagged and quoted before you sign
- Ongoing hygiene: Keeping the vault current handled under standard retainers, from €500/month
Think of it the way you'd think of clean data before analytics: the agents are the visible output, but the brain is what makes them trustworthy. That's why bundling it with an agent build comes with a saving, the foundation and the agents, planned and priced as one.
Where it fits
The layer beneath every tier.
Second Brain isn't a competitor to the agent tiers, it's the ground they stand on. See the full ladder on Services and the org-level map in AI across three organisational levels.
Starting with one agent?
A Spark agent can run on a lightweight vault. If knowledge is your real bottleneck, build the brain first, then Spark on top.
Deploying a team of agents?
For Starter and Pro, the shared vault is what keeps multiple agents consistent. We bundle it in and knock €1,500 off the combined price.
Multi-department rollout?
For Enterprise, the Second Brain becomes the org-wide knowledge backbone, with per-department spaces and access rules.
Already have agents?
Retrofit a brain under an existing fleet to cut hallucinations and stop re-explaining the same context. Pairs with a retainer.
FAQ
Common questions about Second Brain with Obsidian.
Why Obsidian instead of a vector database or Notion?
Obsidian stores knowledge as plain-markdown files you own, no lock-in, no per-seat data tax. Because the files are linked and human-readable, an agent reads and updates them the way a researcher would, and on small-to-medium knowledge bases this wiki-style retrieval can cut token usage dramatically versus naive vector-database RAG. Notion and Confluence are great for humans but lock knowledge in a proprietary database agents reach through rate-limited APIs. With Obsidian the same files serve your team and every agent.
How do agents actually read from and feed the vault?
Agents connect through the Model Context Protocol or a retrieval layer. They search the vault, pull relevant notes into context, and answer grounded in those sources. On write-back, they append summaries, decisions, and resolved items under controlled paths, so every task leaves the brain a little smarter. Humans and agents edit the same files, keeping one source of truth.
What custom skills come with it?
Every build includes a starter set of custom Claude skills that act on your vault, typically a knowledge concierge (answers from the vault with citations, escalates when unsure), an intake skill (turns transcripts, emails, and tickets into clean atomic notes), and a vault keeper (scheduled hygiene). More specialized skills are built as add-ons or under a retainer. Skills run on the open Model Context Protocol and are yours to keep.
Do we own the knowledge base?
Completely. It's a folder of plain-markdown files on your own storage. No proprietary format, no platform subscription to us, nothing to export if you leave. Obsidian itself is free. You can back up, version-control, and move the files anywhere, any markdown tool opens them.
What happens when the knowledge base is wrong or outdated?
The agent's output is wrong, which is exactly why hygiene matters. Keeping the vault current is the most important ongoing task and is handled under a retainer. We rerun the agent eval suite after meaningful knowledge updates so retrieval quality stays verified.
Is this the same as fine-tuning a model?
No. Fine-tuning changes the model itself, expensive, slow, hard to update. A Second Brain changes what the model sees at inference time via RAG. It's the standard 2026 pattern for business-specific knowledge: update a markdown file and every agent is current instantly, no retraining.
Can you build the brain under agents we already have?
Yes. We retrofit a vault beneath an existing fleet (even agents we didn't build) to reduce hallucinations and stop re-explaining the same context every session. We map your current sources, design the taxonomy, ingest, and re-wire retrieval. Usually paired with a retainer for ongoing upkeep.
How much of our knowledge can the vault hold?
We scope the document set with you in discovery and right-size the build to it. The markdown approach is strongest from a handful up to a few hundred well-structured notes, and remember those are curated, deduped notes, not your raw file count, so a large pile of documents often distills into a far smaller, cleaner vault. For very large or unstructured archives we layer in a vector store alongside the vault. If your corpus is unusually large or messy, we flag it and quote it before you sign, no surprises.