Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.eluu.ai/llms.txt

Use this file to discover all available pages before exploring further.

The most important decision when you set up Eluu is how you scope each colleague. Get this right and the platform pays dividends. Get it wrong and you have an expensive chatbot.

One colleague, one role

A colleague maps to a business role, not a task type. The two ways this commonly goes wrong:
  • Role too broad. A “general assistant” that does sales drafting, customer support, and research. Each domain pollutes the others’ memory. Quality drops.
  • Role too narrow. A “weekly Tuesday email digest” colleague. The role is a single output, not a function. The workspace doesn’t accumulate enough context to be useful.
The sweet spot is a recognisable business role — the kind of thing you might hire a human to do, or buy a SaaS product for.
GoodWhy
Senior sales operations colleague managing pipeline reviews and outreachA real, hireable role. The workspace memory will accumulate around accounts, opportunities, and win patterns.
Customer success lead watching for at-risk accounts and drafting renewal outreachA real, hireable role. Memory accumulates around accounts, conversations, signals.
Engineering on-call colleague triaging alerts and drafting incident updatesA real, hireable role. Memory accumulates around services, recurring issues, and decision history.
AvoidWhy
Helpful assistant for everythingToo vague. Memory is shallow because no domain ever gets concentrated context.
Tuesday standup writerToo narrow. The colleague has nothing to memorise between Tuesdays.
Slack and Gmail botTool-shaped, not role-shaped. The actual job is unclear.

Sizing by org

How tightly you scope a role depends on the breadth of the work and the size of the team.

Personal use or 1-person team

One colleague usually covers a whole function. A “sales colleague” might own prospecting, outreach, pipeline tracking, and weekly reporting. The same colleague handles every part of the workflow because there’s only one person doing it. Pattern: one colleague per function you care about. Three to five colleagues total.

Startups (5-25 people)

One colleague per business function. Sales, customer success, ops, marketing, research, finance ops. Each colleague focuses on its function but covers the breadth of work within it. Pattern: one colleague per function. Five to ten colleagues total.

Scaling (25-100 people)

Functions get specialised. Sales splits into “outbound prospecting” and “renewal management”. Customer success splits into “onboarding” and “ongoing relationship”. Operations becomes “meeting intelligence” + “knowledge management” + “process automation”. Pattern: one colleague per sub-function. Ten to thirty colleagues across the org.

Enterprise

Many colleagues running real workloads in parallel, on schedules, with verification loops. Each colleague is narrow enough to be auditable but deep enough to accumulate years of operational memory. Pattern: role-cluster ownership. Dozens of colleagues across the org.

How a colleague spends its time

Once a colleague is in production, its work falls into two stages — most teams pass through both.

Stage 1 — Conversational work

You ask, the colleague does. Research, drafting, organising files, summarising documents, building one-off views. The colleague is reactive — it works when you message it. This is what most people start with. It’s useful from day one, but the colleague’s value caps at how often you remember to ask.

Stage 2 — Automated workflows with verification loops

You convert the most-repeated conversational asks into scheduled jobs. The morning briefing runs at 8am every weekday. The pipeline check runs every two hours. The customer health digest runs every Friday. With verification loops, the colleague catches its own mistakes. It re-runs the work, compares outputs, flags discrepancies, and only ships when the result passes its own checks. This is where Eluu changes from “useful” to “operational”. The hours saved compound. The memory deepens. Switching off would lose real continuity. The progression is the goal: start conversational, end automated.

Splitting and merging

You’ll occasionally hit a colleague that’s grown too broad. Symptoms:
  • Replies feel less sharp than they did.
  • The CLAUDE.md instructions are getting long and contradictory.
  • You’re inserting “in the context of X” qualifiers in every message.
Fix: split it. Create a second colleague. Copy the relevant sections of the original colleague’s instructions into the new one. The two colleagues will start with the same memory but diverge as their workspaces accumulate domain-specific context. Merging happens less often, but it’s the inverse: two colleagues whose roles overlap and confuse teammates can fold into one.

Where to next

Create your first colleague

The five-step wizard for putting these patterns into practice.

Mental model

How colleagues, sessions, views, and jobs fit together.

Personal vs Shared

Visibility model for colleagues and what changes when you flip it.

Schedule a job

Move from conversational to automated.