The agent harness on top of your company brain.
Hive holds what your company knows. Sayya is what works in it: where people ask their questions, and the layer that chooses which tools to use and which model should reason. It isn't tied to one provider.
- Your Claude or ChatGPT seat
- Anthropic
- OpenAI
- Open-weight on Fireworks
- Picks the tools
- Picks the model
- Remembers what it learns
- Drafts actions for you
- People
- Customers
- Deals
- Decisions
- Sources
- Web
- Mobile
- Google Chat
- Telegram
- Claude
- ChatGPT
Hive is the constant. The model in front of it is the variable.
This is the part only the harness does. One request can use several models, each on the work it does best, while the context underneath stays put. Companies often use more than one of these paths at once.
You, on the webPrep me for my Kelvane call tomorrow.
- Low-cost tier · extractionRead
Pulls the 14 emails, 3 meetings and the open renewal that matter for Kelvane.
- Frontier model · synthesisWrite
Builds the brief: where the renewal stands, what Dana asked for, and what is at risk.
- Low-cost tier · reflectionLearn
After the conversation, notes that you want briefs in five bullets with the contract date first.
Illustration. People, companies and numbers are made up.
- Your own seat
Ask from the agreement you already signed.
A question asked from your own Claude or ChatGPT seat runs on that seat, so your negotiated rate and data-retention terms stay as they are. A dedicated install can run on your own OpenAI, Azure OpenAI or Anthropic account.
- Our keys
Several providers, nothing to negotiate.
Object Edge's keys across Anthropic, OpenAI and Fireworks. One line item, and no second model contract for procurement.
- Open-weight models
Open models beside the frontier ones.
Open-weight models, run by Fireworks, take the work they are best suited to, with the same context underneath.
- Mixed routing
Different work, different models, same context.
Quick extraction goes to a low-cost tier, deep synthesis to a frontier model, and reflection after a conversation to a low-cost one again. The context underneath stays put.
What Sayya does once the company brain is in place.
Built for real operating rhythms: daily planning, revenue work, task tracking, decisions, follow-ups and questions that cross teams. Here is one account manager's day.
- 07:40Daily briefings · Web
Start the day with priorities, blockers and decisions waiting on you.
Sayya pulls overnight email, open commitments, initiative risk and pending follow-ups into one briefing before your first meeting.
Your morningTue- Kelvane: Dana asked to hold the price for a two-year term
- 2 decisions waiting on you
- The follow-up you promised Larkspur Freight is due today
- 09:15Questions · Google Chat
Ask the business in plain language and get a sourced answer.
Status, ownership, deal exposure, initiative progress, open risks and past decisions resolve against the company brain, not a pile of search results.
Who owns the Kelvane renewal now?Maya Lindqvist, since 12 March. Tom Ellis had it before.SalesforceGmail · 12 Mar - 10:30CRM updates · Web
Capture deal context without turning reps into data clerks.
In HubSpot, Sayya can create deals and contacts, log calls, emails and meetings, and move deal stages, from the conversations and follow-ups where the work happened.
Ready for HubSpotDraftLog call with Dana Okafor. Move the Kelvane renewal to Negotiation.
- 12:05Tasks and decisions · Telegram
Turn conversations into assigned work and lasting records.
It creates tasks, records decisions with their trade-offs, links work to initiatives, and keeps the record current as talk turns into action.
Decision Hold list price, offer a two-year term. Trade-off: a lower uplift for a longer commitment.Task Send revised terms to Dana. Owner Maya, due Friday. - 14:00Monitoring · On a schedule
Watch for what matters without anyone polling systems.
Heartbeats run on a schedule; signals watch for events you define. Both report to you, and act only within the permissions you grant.
Renewals due in 60 daysWeekly3 renewals due. One has had no owner since its rep moved teams. - 15:30Channels · Claude
The same agent follows you across web, chat and mobile.
On the web, in Telegram and Google Chat, and from Claude and ChatGPT, with the same company context wherever your people already work.
From Claude: what did Dana ask for?A price hold for a two-year term, in her email of 2 October.Gmail · 2 Oct - 16:45Documents and email · Web
Drafts that reflect what is actually happening.
Agendas, follow-up emails, status reports and documents draw on the live company brain rather than templates and memory.
Email to Dana OkaforDraftDana, thanks for the call. As discussed, we can hold the current price on a two-year term. Revised terms attached.
- 17:20Company memory · Web
Keep playbooks, learnings and precedent after people move on.
Ask how a negotiation was handled, what worked in an earlier launch, or what fixed a similar problem, and Sayya finds the context behind the answer.
Have we held price for a longer term before?Yes, twice last year. Both renewed at the end of the term.Decision · JunDecision · Sep
Illustration. People, companies and numbers are made up.
Three layers of memory keep the agent grounded.
The current conversation, the patterns learned from earlier work, and the company brain all shape how Sayya reasons. That is the difference between a chat session and an agent that lasts.
The company brain as ground truth.
Entities, relationships, history and sources let Sayya check what the business knows before it searches wider.
- Kelvane
- Dana Okafor
- Q4 renewal
- Owner: Maya Lindqvist
- Two price holds last year
Preferences, corrections and validated learnings.
Memory for each person and for the company holds what Sayya has learned about communication style, playbooks, workflows that worked and recurring context.
- Briefs in five bullets, contract date first
- Say "renewal", not "re-up"
The live state of the conversation.
What has been asked, what has been found, and which actions are under way.
- Asked: prep for Kelvane
- Found: 14 emails, 3 meetings
- Under way: the brief
Every conversation should make the system more useful.
Sayya reflects after a conversation and stores explicit notes rather than vague personalization, so what it learns stays inspectable and tied to real work.
- 1 · A conversation
"Not that date. Use the contract's end date."
- 2 · Reflection
Afterwards, Sayya works out what it should do differently.
- 3 · A note you can read
Written down plainly, where people can see it.
- 4 · The next answer
Uses the contract's end date, for everyone who asks.
and round again
When someone asks for a renewal date, use the end date in the signed contract, not the close date in the CRM.
CorrectionLearned in a conversationApplies to the whole company- Corrections
Feedback becomes lasting guidance.
When someone says “not that way”, Sayya records the preferred path and follows it next time, instead of relearning the same lesson.
- Preferences
It adapts to how each person wants information.
Length, format, channel, tone, rounding and escalation preferences become part of the working profile.
- Vocabulary
Company language stops needing translation.
Whether your team says campaign, initiative, deck, queue, territory or port call, Sayya learns the terms and uses them consistently.
- What worked
Approaches that worked shape the next answer.
Sayya records which approaches worked, such as a company-brain query beating a broad file search for account questions, and uses those notes next time.
Autonomous doesn't mean unsupervised.
Sayya drafts, and you confirm anything that leaves the company. Hive asks for your confirmation before sending email or changing your CRM from chat, and records what it changed. Scheduled agents act only within the permissions you grant.
- Sayya does
- Surfaces insights, risks and next steps.
- You do
- Take the action yourself, if you want to.
- Sayya does
- Proposes the task, the decision or the change, and waits for you.
- You do
- Accept it or dismiss it.
- Sayya does
- Drafts the email or prepares the CRM change, then waits for your explicit confirmation before it runs.
- You do
- Confirm it. Hive records what it changed.
Dana, as discussed, we can hold the current price on a two-year term. Revised terms attached.
- Sayya does
- A scheduled agent acts on its own only within a permission you granted, such as sending a recurring report from your Gmail.
- You do
- Grant it once. Pause or change it at any time.
Three differences a better model doesn't close.
Does the agent stop at its vendor's boundary?
Assistants are quick inside their own vendor's suite and awkward outside it. Reaching a CRM or a tracker takes extra tooling, and carrying context between surfaces falls back on the person.
Sayya works across Google Workspace, Salesforce, HubSpot, Jira, GitHub and your data warehouse through one context, and that context follows the person across web, chat and mobile.
Does last week's correction survive?
Stateless assistants reset. The preference you stated, the format you asked for and the correction you made are gone by the next session.
Corrections, preferences, vocabulary and approaches that worked persist as lasting, inspectable guidance, for the whole company, not only the person who taught them.
Who decides which model runs your work?
An ecosystem assistant is that vendor's models on that vendor's schedule. When a better model ships elsewhere, the work built around the old one doesn't come with you.
Sayya is model-agnostic by design. Choose among the models Hive supports, send different tiers to different providers, and change them as the market moves. Your context, memory and workflows stay where they are.
Questions and answers
- What is an agent harness?
- The layer between a person and a set of AI models that decides which tools to use, which model should reason about a request, and how the result is put together. Sayya is Hive's harness, built by Object Edge, and it runs on Hive's company brain.
- Is Sayya tied to one AI model?
- No. Sayya is model-agnostic. It runs on Object Edge's keys across Anthropic, OpenAI and Fireworks, and different requests can go to different providers. A dedicated install can run on your own OpenAI, Azure OpenAI or Anthropic account, and a question asked from your own Claude or ChatGPT seat runs on that seat.
- Can we use Sayya without Hive?
- Sayya is built to reason over a company brain, so it comes with Hive. The reverse works: an outside assistant such as Claude or ChatGPT can query Hive directly, but it gives up the orchestration, the memory and the action paths Sayya provides.
- How does Sayya remember things?
- Three layers. Working memory holds the live conversation. Episodic memory keeps validated preferences, corrections and playbooks for each person and for the company. Long-term memory is Hive's company brain, the company's ground truth.
- Can Sayya take actions, or only answer questions?
- It can act. Sayya drafts, and you confirm anything that leaves the company: Hive asks for your confirmation before sending email or changing your CRM from chat, and records what it changed. Scheduled agents act only within the permissions you grant.
One agent layer, every way Hive is bought.
- HiveThe company brain that makes Sayya useful.
Company brain, memory and sources
- PricingOne company brain under every AI seat you have.
Team at $25 per person a month
- AWS MarketplaceBuy Hive through the AWS account you already have.
Billed on your AWS invoice
- Implementation by Object EdgeObject Edge's engineers point Hive at one named problem.
On objectedge.com
- ArchitectureThe knowledge graph and the controls underneath the company brain.
Hosted in the US, or in your own cloud