Labels every reply
Labels each new reply with its intent, then acts on it: a deal, a task and a draft answer for an interested reply; suppression for an unsubscribe; the sequence resumed after an out-of-office.
reply agentSales → AI agentsAI on your key · no token markup
AI agents that work your replies,on the model you choose.
The reply agent reads every reply, labels it and does the follow-through in your CRM, then drafts your answer for you to approve. T-Ex also drafts sequences and personalizes sends. It never skips a plan gate, a usage allowance or the suppression list.
Start freeNo credit card required
What the reply agent doesReply agenton your keyWaiting for your approval
Priya ShahLumen Analytics2 min ago
“Timing works. Can we talk Thursday about rolling this out to the team?”
get_threadthread 3010.7s3 messages · Founders Q4 · after email 2
classify_threadthread 3011.1smeeting_request
create_dealLumen Analytics0.9sOpened in Lead · From reply
create_taskBook the call with Priya0.7sDue tomorrow
draft_replythread 3011.5sDraft waiting for your approval
Labelled it a meeting request, opened a deal and a task, and drafted the answer. Nothing goes out until you approve it.
Runs onClaudeOpenAIGeminiGrokLocal
Illustrative run with example data. Every call is a tool the reply agent runs today, on your workspace’s own key.
What runs today
Six jobs, each where you already work.
No separate AI screen to learn. Each job sits on the page it serves and reads the same workspace records as the rest of the product.
Guards your sending
Reads each mailbox’s and campaign’s bounces over the last 7 days. When they climb, it lowers that mailbox’s daily limit or pauses the mailbox or the campaign, and says why. It never raises a limit.
deliverability agentSales → AI agentsResearches each company
Reads up to four pages of a company’s own website and saves what it finds, each fact with the page and the words it came from, and one opening line for {{research_line}} in your sequences.
research agentCompanies → ResearchKeeps deals current
When someone with an open deal replies, it reads the reply with the deal: it moves the deal to the stage the reply shows or marks it lost with their reason, asking you first, gives a deal with no owner to the person’s owner and files the next step as a task. It never marks a deal won.
pipeline agentSales → AI agentsDrafts a whole sequence
Your ICP and value proposition in; three to six steps with delays, subject lines and merge tags out, ready to edit.
sequence_from_icpSequences → Generate with AIWrites the first email
One line about the goal becomes a sequence name, a subject and a first message.
sequence_starterSequences → Draft the first emailPersonalizes a send
A per-contact rewrite at send time, cached per enrollment. If the model is unavailable or the cap is reached, the template goes out unchanged.
personalizeSendingBring your own model
Models the workspace can attach
One key per workspace, set under Settings → AI & T-Ex. Switch providers without touching a sequence.
Claude
The Anthropic Messages API. Paste the key under Settings → AI & T-Ex. Runs the agents.
OpenAI
Any OpenAI-compatible /v1 endpoint: your OpenAI or Azure base URL. Runs the agents.
Gemini
Through its OpenAI-compatible endpoint, set as the workspace’s base URL.
Grok
xAI at https://api.x.ai/v1, on the same OpenAI-compatible path.
Your own model
vLLM, Ollama or any OpenAI-compatible server on a public https address.
T-Ex Core
The core.texllm.com service with its own tk_ key, for the writing tasks. It has no tool calling, so the agents need one of the others.
MCP server
Your own agents, on the same API keys.
The MCP server gives Claude Code, Cursor, n8n or an agent you build the REST API's operations as tools: find and add contacts, put them in a sequence or a campaign, check mailboxes and status. It runs on your API keys and their scopes, so an assistant can do only what its key allows.
Connect it with one URL and a key from Settings → Access.
LiveConnecting mailboxes, building sequences and campaigns, and the reply inbox stay in the app.
How to connect itGuardrails
What keeps the model in its lane
Each of these is enforced in the product, not promised in a policy.
A person approves what goes out
A drafted reply waits for approval unless you put that intent on autopilot, and the sender sends it like any other email; the model never sends.
Every research line has a source
A fact is kept only when its quote is on a page the research agent read. A line is saved only when it is built on those facts and names nothing they do not say; anyone it has no line for gets your fallback.
Sending only ever goes down
The deliverability agent can lower a mailbox’s limit or pause a mailbox or a campaign. It cannot raise a limit, and what it paused sends again only when a person approves.
Every word on the record
Each text an agent writes is kept with the model, its inputs and who approved it, and every run shows on AI agents and in Runs with what it did.
Your key, encrypted
Stored with AES-256-GCM and only ever shown back masked. It is used for this workspace’s calls and nothing else.
No token markup
You pay your model provider directly. Outboundrix does not resell tokens.
Plan gates and monthly allowances
The agents and the writing tasks need the Teams plan; agent steps and AI runs count against the plan’s monthly allowance.
Rate limited
Every AI endpoint shares a per-workspace rate limit, so a script cannot run up your bill.
Suppression always wins
The sender checks the suppression list before every send, whether a person or the model wrote the email.
Run the motion, not the stack.
Bring your list, send from mailboxes you own, and let the reply agent work the replies. Start on the free plan: no card, no sales call.
Start freeNo credit card required