octa: our agentic GTM foundation model

octa. An agentic foundation model for marketing, GTM, and CRM.

Built on top of a strong open source base and developed in-house. A harness and an LLM work in tandem. octa is built on a decade of authorized campaign data from CIENCE and Tenbound, from 2,500+ B2B brands. New customers' data does not train the model.
01 / The data advantage

A decade of real campaigns, structured for training.

Foundation models are bottlenecked by their training data. The public web teaches them prose. octa is trained on real GTM outcomes.

10+ yrs

of telemetry

Continuous campaign telemetry from 2015 onward. Multi-region, multi-industry, multi-channel.

1,000+

B2B brands

Real accounts across SaaS, finance, healthcare, manufacturing, media, and public sector.

Millions

of touches

Email, voice, SMS, social, chat, and landing pages. Every variant logged with the outcome it earned.

100Ks

labeled replies

Qualified replies tagged with the why behind every conversion and disqualification.

02 / The gap

Why generic LLMs miss GTM.

General-purpose LLMs are excellent at prose. They are not built for GTM. They have never watched a rep handle a real objection, seen which follow-up turns a reply into a meeting, or learned the cadence that beats the alternative in a specific segment.

octa is built the other way around. The model starts from real campaign outcomes, then learns to reason and write on top of that.

Generic LLM

Trained on the public web. Knows what a cold email looks like.

octa

Trained on real campaigns. Learns which cold email got the meeting.

03 / The origin

From a thousand SDRs to one model.

A decade of CIENCE campaigns and years of Tenbound research generated the raw signal. Capchase contributed both archives into graph8 in 2025. graph8 is turning the combined corpus into octa.

  1. 01
    2015

    CIENCE founded

    Outbound campaign service launches. Day one: log every touch, reply, meeting set, and disqualification reason.

  2. 02
    2015 to 2024

    A thousand SDRs, 2,500+ brands

    Daily CIENCE campaigns across SaaS, finance, healthcare, manufacturing, and public sector.

  3. 03
    2025

    graph8 platform launches

    AI sales platform ships with the harness, agents, and integrations. octa is on the roadmap.

  4. 04
    2025

    Capchase becomes a graph8 shareholder

    Capchase joins as a shareholder and contributes both the CIENCE and Tenbound campaign archives into the company.

  5. 05
    2026

    octa training begins

    Pretraining starts on top of a strong open source base, using the combined CIENCE and Tenbound corpus.

  6. 06
    Underway

    octa-mini, octa, octa-reasoning

    Three model sizes in development: low-latency drafting, balanced agentic execution, and deep reasoning.

04 / How octa learns

Ingest. Label. Pre-train. RL. Ship.

Campaign events from the CIENCE and Tenbound archives come in. Human reps tag the why. octa trains. octa Bench scores. New weights ship to every graph8 org. New customers' data does not train the model.

  1. Step 01

    Ingest

    Campaign events from the CIENCE and Tenbound archives come in: sends, opens, replies, calls, meetings, deal stages, sessions, and form submits.

  2. Step 02

    Label

    Human SDRs and AEs tag the why: which lines worked, which objections killed the deal, which page variant converted.

  3. Step 03

    Continued pre-training

    Start with a strong open source base model. Run a large-scale continued pre-training pass on the GTM corpus.

  4. Step 04

    Long-horizon RL

    Rollouts of realistic GTM tasks: write the next message, plan the sequence, qualify the reply, and generate the page.

  5. Step 05

    Evaluate and ship

    octa Bench replays held-out historical campaigns. New weights ship to every graph8 org.

05 / Your data

Your data trains your model only if you opt in.

octa is built on a decade of authorized campaign data from CIENCE and Tenbound. New customers' data does not train the model.

A customer can opt in to train a model for its own workspace. That model learns from the customer's campaigns and stays in the customer's workspace.

01 Opt in 02 Campaigns 03 Label 04 Your model
06 / The lineup

Three model sizes, one foundation.

Same training corpus, three deployment shapes. Latency matters in the inbox. Reasoning matters in the deal review.

octa-mini

Fast . always-on

Small, fast, always-on. Drafts emails, fills sequence variables, and scores reply intent in milliseconds inside the inbox and dialer.

DraftFill variablesScore intent
octa Default

The workhorse

The default workhorse. Plans sequences, qualifies replies, generates landing pages, writes follow-ups, and books meetings.

Plan sequenceQualifyGenerate pageBook
octa-reasoning

Deep work

For complex GTM work: account-level strategy, multi-channel orchestration, objection handling, and deal risk.

StrategyOrchestrateDeal risk
07 / The progress deck

How we are building octa.

Five short slides on the work in progress: strong open source base, continued pre-training, long-horizon RL, internal benchmark, ship.

  1. 01

    Deep GTM knowledge. Hard tasks to completion. Realistic motions.

    octa is being built around three goals: real domain knowledge, the ability to run hard GTM tasks to a booked meeting or generated landing page, and evaluation on realistic motions our customers run every day.

  2. 02

    Continued pre-training on a strong open source base.

    We run a large-scale continued pre-training phase on the CIENCE and Tenbound corpus to push the model toward the specific work it will do inside GTM teams.

  3. 03

    Rollouts of real campaigns. Length penalty. Self-summarization.

    Long-horizon reinforcement learning collects realistic GTM problems, runs many rollouts per problem, scores them on real outcomes, and updates toward winners.

  4. 04

    octa Bench uses real anonymized campaign queries.

    Public LLM benchmarks measure trivia and code puzzles. octa Bench is built from anonymized real queries pulled from the corpus.

  5. 05

    octa-mini today, octa next, octa-reasoning soon.

    The first model sizes target fast drafting, balanced agentic execution, and deep reasoning. We will publish numbers when they survive evaluation.

08 / Goals

What we are optimizing octa for.

We will not publish benchmarks we have not earned. These are the outcomes octa is measured on.

01

Reply rate on cold outbound

Raise the reply rate on held-out campaigns from real industries, personas, and channels.

02

Meeting-set rate from positive reply

Convert inbound interest into calendar invites. The next three lines after a positive reply are where many teams lose the meeting.

03

Qualification accuracy

Route fit vs. non-fit replies against ICP, budget, timing, and authority before a human spends a minute.

04

Landing page conversion

Generate pages that convert clicks into pipeline: layout, headline, form, and proof.

09 / Side by side

octa vs general LLMs, by GTM task.

Same input, different upbringing. Generic LLMs read the public web. octa is built on a decade of campaigns.

Cold email

Picks the opener that this persona actually replied to in historical campaigns.

Writes plausible prose with no link to reply rate.

Qualification reply

Recognizes the objection pattern and drafts the response that reopened similar conversations.

Apologizes generically.

Sequence plan

Knows which cadence beat the alternative in this segment and which channel to switch to mid-sequence.

Guesses a generic 5-touch cadence.

Landing page

Generates the layout, headline, form, and proof block grounded in what converted for similar offers.

Returns copy, no page.

10 / octa

Six specialized heads, one model.

octa is not a single chat model wrapped in a prompt. Each GTM job has its own head trained on the slice of corpus that matters.

Research

Builds account briefs, persona docs, and competitor maps from the open web plus your private context.

Copy

Drafts emails, ads, and sequences in your brand voice, scored against historical reply rates.

Pages

Generates full landing pages: layout, headline, form, and proof. Not just text.

Qualify

Reads inbound replies and meeting transcripts. Scores fit and routes to the right rep with context.

Coach

Reviews call recordings and reply threads. Surfaces what worked and gives reps the next best action.

Deal risk

Flags deals at risk and stalled deals.

11 / Today

What runs today, what is training.

The octa harness is live inside graph8. octa's own weights are in training on the CIENCE and Tenbound corpus. New customers' data does not train the model.

24 / 7

The octa pipeline runs inside graph8: drafting, generating, and qualifying on a strong open source base while octa's own weights train.

6 heads

Each specialized on one GTM job.

Opt in

New customers' data does not train octa. A customer can opt in to train a model for its own workspace.

07 / Pricing

You pay for what the agents run. Every user is included on the Team plan.

Execution credits send, call, and run agents. Contact credits reveal and enrich the people you target. Three ways in.

For trying it out Pay as you go
Free to start

A free entry point to the full platform. No credit card.

  • 1,000 contact reveals to start
  • 500 execution actions free
  • Full platform access, no card
  • Then $0.05 per credit
Start Free
For data in your AI tools MCP Unlimited
$25 / user

Unlimited contact data, in the tools you already use.

  • Unlimited contact data, fair use
  • 2,500 execution credits a month
  • Priced per user, not per org
Claude ChatGPT Cursor VS Code
Start Free
Recommended
The full platform Team plan
$99 / mo

Unlimited users and data. Every app, one bill.

  • Campaigns, dialer, AI voice, and AI workflows
  • Meetings routed and booked for you
  • 10,000 execution credits a month
  • Unlimited users, no seat fees
  • 7-day free trial, no credit card
Start Free

Everything in the Team plan

$99 / mo, all included

The $99 Team plan is the full platform: everything below, for the whole org. MCP Unlimited and Pay as you go are lighter entry points with usage limits.

What the agents run

10,000 credits included
  • Campaigns and sending infrastructure Mailboxes, domains, and deliverability run for you, end to end.
  • 5-channel sequencer Email, phone, LinkedIn, SMS, and WhatsApp in automated sequences.
  • Dialer and AI voice agents Reps call from the built-in dialer. AI voice answers inbound calls, makes outbound ones and books the meeting.
  • AI inbox Every inbound email, LinkedIn message and reply sorted, drafted and answered.
  • Meeting routing and booking Route leads to the right rep and book the meeting automatically.
  • CRM, pipeline, and quotes Deals, quotes, and each rep's daily list in My Desk.
  • Studio content engine Brand, newsletters, landing pages, ads, and campaign content.
  • AI agents and workflows Operator and Executor agents for RevOps, deal reviews, and SDR work.

Contact data

Included
  • MCP Unlimited for every user A $25 per user value, included for the whole org on the Team plan.
  • No monthly cap on contact data Reveals, lookups, enrichments, and exports for every user.
  • Buying signals Visitors, intent, hiring, and job-change signals on every account.
  • Waterfall enrichment Third-party providers, pay on match or bring your own keys.
  • Free search and browse Your whole team explores 700M+ contacts and 100M+ companies.

For the whole team

Always on
  • Unlimited users, no seat fees Invite everyone. Pricing never scales with headcount.
  • Every feature unlocked No tiers gating voice, enrichment, sequencer, campaigns, or graph8 Desktop.
  • Built-in CDP and integrations Warehouse, 500+ connectors, two-way CRM sync included.
  • Month-to-month, no contracts Cancel anytime.
  • Fair-use rate limits Limits apply on all plans. They guard against scraping, not real GTM work.

Execution credits send, call, and run agents. Contact credits reveal and enrich the people you target.

Data tax ROI calculator
octa model

Bring your motion. octa already knows the patterns.

Replay one of your past campaigns through octa and see the lines, cadence, landing page, and meetings the model would have produced.

Start Free Read octa squared