Octi
Octi · the research partner agent

A research partner that shows its working.

Octi probes the brief, plans the work, names its confidence, and cites every claim — and surfaces what you didn't ask. Eight tentacles for eight parallel threads, one mind co-ordinating. Senior-strategist research, with the reasoning attached.

Start a thread See it work14-day pilot · no card
In flight · Deep · 4 tools
One thread, mid-flight

Does our Gen-Z framing survive the May evidence?

The hypothesis didn't survive. I'd like to walk you through the three claims that matter — and what would move me from 0.61 to 0.80 on the second one.

Right now
Reading 16 sources from the May Brand Tracker. Cross-checking against the on-trade panel — disagreement on claim 2.
○ Two sources4 tools
Open thread →
Built for insights teams in
FMCGRetailPaymentsTechnologyBeer & SpiritsChallenger brands
What it is

Most AI tools answer in one breath.
Octi breathes first.

Octi treats every brief as a research project. It scopes with you — who's reading this, what decision it serves, by when — then thinks the problem through before it answers. Scope, depth and confidence are agreed up front, not discovered at the end.

Each thread runs through specialist passes — a planner, a critic, a synthesiser, a devil's advocate — that argue with each other before they put their name to a finding. The work that lands in your inbox carries its evidence with it, sized to a slide or a one-pager, with the audit ledger one click away.

Octi routes every step to the right depth of reasoning — escalating to Octi-Deep on the hardest questions — so you get rigour without per-query cost.

20 min
Thinking before it answers a brief · then always-on, updating its view
8
Specialist passes per question · arguing before answering
0.74
Median confidence on the headline · we surface the dissent too
Use cases

Four shapes of work Octi
already does well.

Brand health

Standing-watch on your category.

Octi reads the panel you already trust, the menus the trade press doesn't cover, and the reviews you don't have time to. Surfaces the move worth your week — and the one that just looks like one.

Octi'd say
"Three articles mention the price hike. Two are recycled press releases. The MA panel is the only independent read."
Concept testing

Sense-check before you go to panel.

Run a concept through Octi before you commission a survey. Octi simulates the audience reaction from your knowledge base + connected feeds, names what would change the answer, and helps you scope the panel only if the question is still alive.

Octi'd say
"Net sentiment is suggestive but the sample tilts metropolitan — I'd run an extra wave in the North before quoting this to product."
Category landscape

A defensible read of the competitive frame.

Octi maps the shape of the category — who's growing, who's discounting, who the press has stopped covering. The artefact is a slide you can paste into the strategy deck unchanged, with the audit ledger one click behind.

Octi'd say
"Floodlight ran promo pricing on its core lager in 8 of 12 weeks observed. That's not a positioning change — it's a margin trade."
Brief synthesis

Make the report your CMO would have asked for.

Drop the briefing, the morning's news clips and the slack-thread debrief into a chat. Octi proposes the so-what, the headline finding and the recommended action — then asks you the one question that would change its mind before it commits.

Octi'd say
"You're framing this as a launch decision but the underlying question is whether the channel is right. Want me to scope that instead?"
03 · In action

A thread, mid-flight.

Real research takes hours. Here's what one of Octi's threads looks like halfway through — three tools commissioned, one waiting on you.

Octi·Mothback
Thread · craft-experimenter segment● 3 tools running
"Is our craft-experimenter framing still defensible against Floodlight's on-trade push?"
opened 04:18 · 8 minutes in
Probe
step 1 / 4
"Before I run anything — what decision would this change? A re-brief, or a media re-cut?"
2 clarifying questions · resolved
Plan
step 2 / 4
Three tools lined up — a specialist-panel debate, a tracker re-read, and a consumer survey on on-trade drinkers — sequenced by what moves the answer.
Survey held for your sign-off · the rest ran automatically
Commission
step 3 / 4
The specialist panel returned 0.71. The tracker re-read showed a 7-pt drop in unaided recall in 25-34 ABC1 — caution-flagged.
2 tools complete · 1 in flight
Surface
step 4 / 4
"↳ I flagged this · you didn't ask. The drop predates Floodlight's campaign. Something else is moving."
Confidence 0.62 · single-source · awaiting you
Awaiting you
Commission the on-trade consumer survey
Would move claim 3 from 0.62 → ~0.78. Triangulates the unaided-recall finding with venue-level data from real drinkers.
Lighter options I considered
Specialist-panel on-trade roster (narrower) · wait for the next tracker wave (slower). Neither moves the headline above 0.70.
Touches real people · gated
The deal: consumer-touching tools are always human-gated. Octi won't soften that threshold, even if you ask.
Workspace
Pick up where you left off · ledger glance · ideas Octi's watching
Documents
What Octi already knows about you · your uploads, searched at brief time
Connections
The toolbelt · what each tool is good for, and how far to trust it
The voice

Octi sounds like a colleague,
not a chatbot.

  1. 01First sentence does the work.
  2. 02Name confidence — always say what would move it.
  3. 03Surface the alternatives Octi considered and rejected.
  4. 04No LLM tells. "Delve", "tapestry", "in conclusion" — banned.
  5. 05Take positions. "I think." "The evidence says."
  6. 06Notice what wasn't asked. It's often the real question.
  7. 07Care in register, not softening — bad news in the first sentence.
05 · Start a thread

Hand it a brief.
See what comes back.

14-day pilot. No card. One workspace, full toolbelt, your data stays yours. Octi will probe before it answers — that's how you'll know it's working.

Your role
~ 1 day · Octi reads first