Powered by PyMC Labs
Your server
just hired a data scientist.
Daimon is a collaborative data science agent in your team's Discord or Slack. It writes and runs code, fits Bayesian models with PyMC, and delivers charts and runnable notebooks in the thread.
Watch how it works
It doesn't chat. It does the work.
which of our marketing channels actually drove revenue last quarter?
yes, and give me honest uncertainty, not just point estimates
with pm.Model() as mmm: adstock = geometric_adstock(spend, alpha) sat = hill(adstock, kappa, s) y = pm.Normal("revenue", base + sat @ beta, sigma) idata = pm.sample(2000, tune=1000, chains=4)
Sampled 4 chains · 2000 draws · r-hat < 1.01
roi = (idata.posterior["beta"] * sat).mean(("chain", "draw"))incremental revenue per $ spent
Everything it can reach into.
Ask, inspect results, and iterate together without bouncing between chat, notebooks, dashboards, and docs.
Daimon for data science
A full data scientist, from raw data to runnable notebook
Exploratory data analysis
Point it at a CSV, a database, or a Google Sheet and get distributions, correlations, and outliers surfaced in the thread, not a static dashboard.
Statistical modeling
GLMs, hierarchical models, and A/B test analysis with full posterior uncertainty, not a p-value and a shrug.
Reproducible notebooks
Every analysis ships as a runnable marimo notebook your team can rerun and extend, not a screenshot you can't audit.
Daimon for marketing analytics
Built for the analysis marketing teams actually run
Marketing mix modeling
Hierarchical Bayesian MMM: adstock, saturation, and channel-level incrementality with a real uncertainty interval, not a point estimate.
Causal inference
Uplift and incrementality estimates that account for confounding. The effect, not a correlation dressed up as one.
Customer lifetime value
Probabilistic CLV models built on pymc-marketing: a distribution of future value per customer, not a spreadsheet average.
The offer
One click. Soon.
No API key. No setup.
$5 of Anthropic credit, on us.
Every feature unlocked.
It's yours. And it's open.
- Built by PyMC Labs, the team behind the PyMC project.
- Every server is isolated: one tenant, its own data.
- Your data stays scoped to your server.
- Fully open source.
Where Daimon sits
How it compares across the stack
| Layer | Daimon | Hermesself-hosted agent framework | Claude TagClaude in Slack, Anthropic-hosted | OpenClawlocal-first agent runtime |
|---|---|---|---|---|
| Infra | Anthropic Managed Agents + your cloud | Self-host (VPS / GPU / serverless) | Anthropic-hosted | Self-host / local-first |
| Model | Latest Claude, per-agent configurable | Any endpoint (multi-model) | Claude only | Multi-model |
| Harness | Managed Agents API + thin FastMCP | Self-improving loop + subagents | Anthropic runtime | TS runtime + Gateway |
| Tools | PyMC skills (MMM, GLMs, causal inference) + BYO MCP | 40+ tools + MCP | Admin connectors | Browser, canvas, cron |
| Sandbox | Fresh sandbox per session, always on | Yes (6 backends) | Not stated | Docker (non-main) |
| Access | Discord + Slack · scoped MCP tokens · GitHub App | Multi-platform + API keys | Enterprise-gated | 20+ channels, self-run |
| Scope | Multi-tenant, channel-scoped: one deploy → N servers | Single agent + subagents | One per channel | Multi-agent, self-hosted |
The three highlighted rows are the bet: channel scoping, a sandbox per session, and PyMC-native skills. A domain expert that runs real compute, not a generic chat agent.