Self-hosted AI workforce
Pre-configured AI hires that handle real work in Gmail, Slack, HubSpot, and 16 more tools. Run on your own infrastructure with one Docker command.
MIT-licensed core · No per-token fees · Bring your own LLM
Each role ships with a job description, a set of skills, and a heartbeat — recurring work it does on its own. Install one. Or install a whole team.
Prospects, researches, and initiates outbound to book meetings.
First-response triage, resolution of known issues, knowledge-base lookups.
Invoicing, dunning, refund processing, failed-payment recovery.
Scheduling and inbox triage — proposes times, books meetings, drafts replies.
Briefs, memos, competitor teardowns, charted findings.
Coordinates a delegation team — hands UI to Designer, planning to TPM.
Translates briefs into concrete UI/UX specs. Pairs with Product.
Turns briefs into sprint plans and RFC outlines. Pairs with Product.
Bundle three at once with Team Packs — pre-wired delegation between agents.
No new tooling to learn. No data leaving your network. No per-token surprises.
One Docker run. Pick local Ollama, OpenRouter, or any OpenAI-compatible endpoint.
docker run -p 3100:3100 \
inhouse/agents:latestOAuth into Gmail, Slack, HubSpot, Salesforce. Or paste an API key for the rest. Capability slots map to your stack.
Choose a Role or a Team Pack. Agents come online with delegation pre-wired and start their heartbeat work immediately.
Everything you need to run a working AI workforce in production — without giving up control.
Pre-built agent templates with system prompts, tools, and heartbeats. Install in seconds.
Multi-agent bundles with delegation pre-wired. One click installs the whole org chart.
Agents declare abstract needs (CRM, email send, scheduling). Bind to whatever tools you already use.
BYO-app OAuth for Google, Microsoft, Salesforce, Slack, HubSpot. Tokens encrypted on disk.
Webhook in, agent fires. Wire any vendor (Stripe, Linear, Zendesk) with templated prompts.
Per-agent token + cost tracking. See who's doing what, where the time went, and what it cost.
Closed AI platforms work for hobbyists. Real ops teams need privacy, predictable cost, and control over what their staff can do.
Your customer data never leaves your network. Run on local Ollama; the only outbound traffic is to the SaaS apps your agents already touch.
Local models cost $0 per token after the GPU. Or use OpenRouter and pay wholesale. No usage-based platform fees on top.
Edit the system prompt. Swap the model. Strip a tool from a channel. Audit every action via per-agent activity logs.
15 minutes. We'll walk through the install and answer your hardest "but what about…" question.
Book a 15-min demo →Or sign in to a hosted instance at app.inhouseagents.ai