zenityaa teammate, not a chatbot
The agent that works where your team already talks.
Bring it into the channel where work already happens. It learns how your team does the job, creates tools it needs, and keeps the repeatable parts running.
SlackDiscordOther
Slack
DiscordJacob09:41
@zenitya we have no way to pull ad numbers. can you build yourself a tool for google ads and meta ads?
Bring the model you already trust.
You are not locked into a single provider. Use any model you want, pick default for entire workspace or specific per channel, and swap it any time.
Your company teaches it how to work.
It learns from the people who know the process, builds the tools it is missing, and turns a good result into work it can repeat in the background.
It builds the tool it needs
Custom tools connect it to anything, including the legacy and proprietary systems modern protocols do not reach.
- Not limited to MCP: connect modern APIs, legacy software, or proprietary systems
- One tool can collect data, analyse it, recommend actions, and output a PDF
- Deterministic execution means better results, fewer tokens, and lower cost
It learns how your company works
Pull the process out of one person's head, in the same conversation where the work happens.
- Ask it to grill the teammate who knows the edge cases
- It turns the thread into a reusable way of working
- The next person gets Sarah's method, not a generic answer
Make it run on schedule
Tell it what to run and when. It takes the work from there, even when nobody is waiting in the thread.
- Turn any learned process into recurring background work
- Review the task and cadence before anything is created
- The result returns to the channel when the work is done
It learns the business between the lines
The details that keep work moving rarely live in a handbook. It remembers the exceptions, commitments, and working habits your team reveals in conversation.
- Carry customer commitments and operating context into the next decision
- Every detail links back to the conversation that taught it
- People can read, correct, or remove what it learned
Developers own the tools
Let developers and technical teammates own the custom-tool code.
- Version tools with Git in your organization’s repositories, and deploy only final versions
- Build with any AI tool or IDE: Claude Code, OpenCode, Cursor, Codex, Grok, or whatever your team prefers
You should always know what it did.
Any model
Anthropic, OpenAI, Gemini, OpenRouter, or self-hosted, your keys.
Per-channel config
Model, instructions, tools, and memory.
Write approvals
State changes show the payload and wait.
Mention-only
Silent by default, ambient replies opt-in.
Audit rows
Who asked, which skill, tools, model, tokens.
Scoped access
Assign tools by workspace, channel, or named user.
MCP server
Your reviewed tools in any MCP client.
Skill sync
Point at a GitHub folder and keep it current.
CLI and SDK
Author tools in your IDE, deploy from a branch.
Spend visibility
Token cost per channel and skill.
Tool versioning
Edits return to draft with a source snapshot.
Report handoff
Long answers become an Artifacts link.
One agent, scoped to every team and person.
Set workspace defaults, give each channel its own context and tools, then reserve sensitive capabilities for specific signed-in users. A get-financials tool can stay CEO/CFO-only, even in a shared channel.
| setting | workspace default | per channel | per user | what it means |
|---|---|---|---|---|
| Model and provider | yes | yes | no | Anthropic, OpenAI, Gemini, OpenRouter, or self-hosted, with its own key. |
| Instructions | yes | yes | no | Place-specific system prompt; the channel's is appended to the org's. |
| Skills | yes | yes | no | Global skills are visible everywhere; channel skills stay in their channel. |
| Tools and connections | yes | yes | yes | Keep get-financials available only to named users, such as the CEO and CFO. |
| Memory | yes | yes | no | Durable facts follow the scope. Readable, editable, and deletable. |
| Schedules | no | yes | no | A schedule runs in the channel that owns it, with that channel's runtime. |
| Response policy | yes | yes | no | Mention-only by default; keyword-gated ambient replies where you want them. |
| Write approvals | yes | no | no | State-changing tool calls always show their payload and wait for a human. |
Teach it once. Let it keep working.
Show it how your business handles one recurring job. It learns the process, uses the right tools, and runs it again in the background.
What teams ask first.
What is Zenitya Agent?
An agent that learns how your business works and keeps useful work running. Bring it into a channel and it can use your approved tools, learned processes, schedules, and memory while people stay in control of access and writes.
How is it different from a Slack chatbot?
A chatbot answers a message. Zenitya Agent carries workspace state: reusable skills, scheduled runs, per-channel memory, sandboxed tools, and an audit row per turn. It is configured like a system, not prompted like a toy.
What is a skill, and who creates it?
A reusable instruction set the agent loads on demand. Write one in the console, sync a folder from GitHub, or ask the agent to save what it just did. Skills are global or scoped to a single channel.
Can it schedule its own work?
Yes. Ask in the thread and it proposes the cron and the skill. Creating the schedule is a write, so it shows the payload and waits for approval. Each run posts back in the same channel.
What does it remember, and can we delete it?
Facts learned in a channel stay in that channel, with a link to the message they came from. Every entry is listed in the console and can be edited or deleted.
Is it safe to let an agent write its own tools?
The draft is the boundary. Agent-written tools never execute until a human reads the source, binds the declared secrets, and enables them. Execution happens in a container with a read-only filesystem and no network unless the reviewed manifest asks for it.
Which models can we use?
Anthropic, OpenAI, Gemini, OpenRouter, or a self-hosted / open-weight endpoint, with your own keys. Set a workspace default and override it per channel or per skill.
How do you stop it reaching the wrong systems?
Access can be assigned at the workspace or channel level, then narrowed to named console users. A tool such as get-financials can remain available only to the CEO and CFO, even inside a shared channel.
Does it respond to every message?
No. Mention-only by default, including inside threads it is already part of. Ambient replies can be enabled per channel behind keyword rules.
Does it work in Discord or Teams?
Slack, Discord, and other chat surfaces are adapters over one runtime. Each surface inherits the same skills, schedules, memory, tools, and access rules.