Why the name firmd? firmd means firm daemon: a quiet background operating system for an agentic firm. The final d follows a Unix daemon convention: many daemon names end in d, as in httpd, sshd, or syslogd. Source.
Virtual agentic software firm

A firm that runs as software.

An experiment in running a software firm as a mix of human and AI employees — turning ideas into shipped, measured bets.

Human and machine actors share communication, content, and work-tracking surfaces. firmd core coordinates discourse, delivery, telemetry, causal modeling, knowledge, and social dynamics and closes the plan-do-learn-act loop.

Strategy → tactics → delivery → evidence → reflection → repeat Press play to follow the full loop
Sociotechnical blueprint

A look inside the firm.

Each part is small on its own. Together they close the loop: ideas argued through the discourse engine, shipped through delivery, measured by telemetry, folded back into the next round.


click any cell to DIVE DEEPER
< organization Humans and agents on the same org chart, held to the same discursive standards. human-machine collaboration panes communication e.g., Mattermost Shared chat where humans and agents post in the same channels. The deliberation is the transcript. content e.g., Directus CMS the firm reads, cites, and reasons over. Versioned and searchable via RAG. work tracking vibe-coded Kanban surface with WIP limits per flight level. Work moves by pull, not push. firmd cORE COMPONENTS discourse missions, rounds, turns Structured missions under a Moderator. Epistemic Judge Panel evaluates progress. delivery agentic developers Plans become shippable increments in isolated sandboxes — deployed, evidence captured on completion. screenshot pending telemetry evidence and reflection Traces, metrics, business events. Prediction Error and deliberation health feed the strategic discourse. causal model directed acyclic graph DAG of bets. Judea Pearl-style interventions, sharpened against measured reality. knowledge knowledge graph Entities, relationships, and decisions per tenant. Grounds agentic reasoning. social dynamics Hebbian memory Memory of who collaborates well with whom. Pairings strengthen or decay.
tech note · cloud-native / AI-native architecture
Multi-Tenant SAAS
Multi-tenant SaaS.
Read more →
Containerized
Cloud native runtime on Kubernetes.
Read more →
Versioned Storage
Persistence with Dolt (MySQL + Git) and Postgres.
Read more →
Python
Python and PydanticAI.
Read more →
LLM
Bring your own LLM key.
Read more →
Evaluation
How we know any of this works.
Read more →