An AI agent in your Slack channels
Not another doc-RAG chatbot. An agent that lives in your channels and gets better every week from your team's feedback.
AI is the easy part
"AI over your company knowledge" is a commodity now — a dozen vendors sell it. The part they skip is the constraint that actually makes it compound: the agent is public by default. It refuses DMs. It lives only in channels. Every answer happens where the whole team can see it, correct it, and learn from it.
That single rule turns the agent into a Lehrwerkstatt — a teaching workshop. When it gets something wrong, someone on the team corrects it in the channel with the right answer. Everyone else sees it in the moment and the agent absorbs it. The knowledge base stops being a graveyard of outdated docs and becomes a living thing the team maintains because it lives in the flow of work.
Generic vendors let you scope which channels the agent listens in, and stop there. They don't enforce channels-only (never DMs), learning-in-the-open, or the ritual of capturing what it missed. Agora is opinionated about exactly the parts they leave out.
What gets installed
A single loop. The agent answers in public, the team curates per-channel zones, gaps get captured, and the skills compound week over week.
Public-by-default agent
"Ask {Client}" refuses DMs and politely asks you to open a public channel. The defining constraint — trivial code, big philosophical statement.
Channel zones
3–5 channels modelled with your team, each with its own zone, skills, and pinned instructions. Read-only into Notion, Drive, GitHub, Linear.
Answers in the open
The team asks in public. Every answer is visible, correctable, and reusable — the collective learning generic doc-RAG never gets.
"What should it have known?"
A reaction or slash command logs every gap into a backlog. Humans fill it. The agent's skills compound with each iteration.
The Lehrwerkstatt loop: Step 4 feeds back into Step 2. Every week the agent knows more, because your people wrote it down.
Who it's for
A fit if you're…
- A growth-stage Latam tech org, roughly 30–150 people in product and engineering.
- Slack-native (or Teams-native) — public channels are already where the work happens.
- Post-Series-A fintech or B2B SaaS, with knowledge sprawling across tools faster than anyone can document it.
- Led by a CTO, Head of Engineering, or COO willing to make working-in-public the default.
Probably not a fit if…
- You're a 5–30 person team. Too small for the collective learning to matter — a shared doc still works.
- Your culture runs on DMs and private threads. Public-by-default would create friction with the culture, not help it.
- You want write actions — PRs, deploys, tickets — on day one. That's a v2 conversation, not this install.
- You need SOC2 paperwork before a pilot. I'm not the lowest-friction vendor for that yet.
Three staged SKUs
Flat fees, never hourly — every efficiency the kernel earns accrues to you. Start with the PoC; from there the natural path is Build and Coaching.
PoC
US$4K–6K
flat · 2 weeks
A live agent in one channel, in two weeks. Plus the diagnosis to scale it to the rest of the team.
- Agent deployed in 1 pilot channel
- Knowledge-sprawl & channel-hygiene audit
- Prioritized rollout plan to the rest of the team
Build
US$20K–35K
flat · 4–8 weeks
The "Ask {Client}" agent, deployed. Public-by-default, channel zones, and the capture flow — live in your workspace.
- DM-refusing, public-only agent
- 3–5 channel zones, skills & instructions
- "What should it have known?" capture flow
- Read-only Notion / Drive / GitHub / Linear
Coaching
US$3K–5K/mo
monthly retainer
So the agent keeps delivering value month over month. I run the Lehrwerkstatt ritual, curate skills, and review the metrics that matter.
- Lehrwerkstatt ritual leadership
- Ongoing skill & instruction curation
- Monthly metrics review
A project that goes through all three SKUs is roughly US$45K–80K in year one. I won't sell a Build without a coaching commitment, because without the weekly review ritual the agent decays — and a decayed agent is a worse outcome than no agent.
Why me
I've shipped AI assistants and agents to clients across US & Latam: Whyspr at MateBio (GraphRAG over a biomedical knowledge graph of 45M+ nodes), an AI chat at a Seattle FinOps startup via Xogito (multi-provider agent, MVP in a month), and the "Ask Magoya" internal agent pattern this offering generalizes.
Each deployment wears your brand — "Ask {YourCompany}" — never Agora's. The agent should feel native to your workspace, not like a vendor bolted on.
Want to see how I design AI architectures and ship agents to production? See Fractional CTO for GenAI.
One design partner slot at half price
Roughly US$10K–15K flat for the Build — about half the eventual price — in exchange for case-study rights and a written testimonial. A design partner whose team is ready for the agent to live in their channels, not DMs. If that's you, let's talk.