The same fixes, every week
You explain the format, the tone, the sources you trust. The chat ends, and the next one starts from zero.
AI workspace for small and mid-size teams
Correct it once. Your whole team gets it right next time.
backlayer is a shared workspace where your team works with Claude or Codex on real work. When people keep fixing the same thing, it drafts a written procedure. Someone on your team approves it, and the agent follows it for everyone.
- Runs on your server
- Your team's Claude or Codex accounts
- A person approves every change
Which accounts renew before December 31 and haven't heard from us in 30 days? Draft a short follow-up for each.
Used procedure · renewal-outreach
Searched Knowledge Base · 3 documents
Four accounts renew before year end with no contact in the last 30 days. I've drafted a follow-up for each, in the tone from your Renewal playbook.
AccountRenewsDraft
Brightline LogisticsNov 14Drafted
Acme Dental GroupNov 30Drafted
Northfield BakeryDec 8Needs review
Carver & SonsDec 19Drafted
Sound familiar?
Everyone on the team uses AI. Nobody's lessons stick.
The owner is the bottleneck
The person who knows what good looks like ends up checking every draft, or maintaining one giant prompt for everyone.
Memory you can't check
When an assistant remembers things on its own, it's hard to see what it learned, where from, and how to undo it.
Use cases
What owners use it for
You know how the work should be done. backlayer turns that into procedures your team and the agent can follow, so the work doesn't wait for you.
Onboard new hires faster
A new hire gets an agent that already knows your templates, your clients and how you write. They ask it first, and you answer the questions that are actually new.
Hand off the repetitive work
Weekly reports, status updates, data clean-up. Approve the procedure once and the agent drafts it the same way every time. Your people review instead of retyping.
Get know-how out of your head
How you price a job, scope a project or write a proposal becomes a written procedure. Your team can read it, improve it and use it without asking you.
Deliver the same quality every time
Client briefs, comparisons and research reports follow the same structure and the same sources, whoever on the team runs them.
Answer customers the same way
Replies follow your policies and your tone, with the facts taken from your Knowledge Base, not from whoever happens to be on shift.
Keep going through vacations and turnover
When someone is out or leaves, what they figured out stays in the Knowledge Base as procedures, not buried in their chat history.
How it works
From a correction to a team procedure, in four steps
- 1
Work as usual
Chat or talk with the agent in your browser. It reads and writes your team's Knowledge Base, so notes and drafts land where everyone can find them.
- 2
Ask for a review
Pick the sessions worth learning from. backlayer reads them and drafts changes to your procedures, in plain text.
- 3
Approve what's right
Each suggestion shows the change, the sessions it came from and why. Approve it, or say no and nothing changes. Every decision is logged.
- 4
Share what works
When a procedure proves itself, an editor shares it with the whole team. Every version is kept, so any change can be rolled back.
Suggested update
client-brief
Learned from 3 of Maria's sessions this week
## Structure- Start with background on the client.+ Open with the decision the client has to make, in one sentence.+ Keep background under 120 words.+ Cite the source for every number.
Why: in all three briefs the decision was moved to the top, and twice someone asked where a figure came from.
Straight answers
What we promise, and what we don't
Your data stays on your server
The workspace, the Knowledge Base, the database and session transcripts live on infrastructure you control.
The model runs at the provider
Prompts and responses go to Anthropic or OpenAI under your own account's terms. We won't pretend otherwise.
Approval isn't proof
Approving a change authorises it; it doesn't prove it helps. That's why we measure corrections and review time, before and after.
Not a coding tool
It's built for research, writing, analysis and operations. Nobody on your team has to touch code.
The pilot
Bring one piece of work you're tired of fixing
We're onboarding a small number of teams of 5 to 50 people. Tell us what you keep correcting or explaining, and we'll agree the setup, scope and timeline with you when we talk.
Questions
Do we need Claude or ChatGPT subscriptions?
Yes. Each person signs in with their own account, directly with Anthropic or OpenAI. The team doesn't share one key.
Where does our data live?
On your server: the workspace, the Knowledge Base, the database and session transcripts. Model inference happens at Anthropic or OpenAI.
Will the agent change its behaviour on its own?
No. Reviews run when you ask, and they only suggest. Nothing changes until someone approves, and a procedure reaches the team only when an editor shares it.
Can a new hire use it on day one?
Yes, once they have an account. They work with the same Knowledge Base and the same shared procedures as everyone else on the team.
Is this a coding tool?
No. Under the hood the agent is Claude Code or Codex, but backlayer is built for knowledge work: research, writing, analysis and operations.
What does it cost?
Pilot terms are agreed per team. Tell us about your team in the form and we'll come back with details.