Organizational awareness through stigmergic signal processing. Surface structural patterns across your engineering tools without anyone having to ask.
Ingest work artifacts. Route through a self-organizing mesh. Surface patterns no single tool reveals.
GitHub PRs, issues, commits, and reviews. Linear issues and cycles. Slack threads and reactions. Grafana metrics and traces. All normalized into a unified signal format.
adaptersSignals route through a competitive mesh using Adaptive Resonance Theory. Workers self-organize into stable categories. New patterns fork new workers; stale ones decay. No retraining.
meshDetect coordination gaps between teams, knowledge silos forming around individuals, and dependency risks that span multiple tools and time horizons.
correlationUnify people across GitHub handles, Slack names, Linear UUIDs, and email addresses. Learns new aliases at runtime. Privacy-first with configurable sensitivity levels.
identityAll output passes through a constraint engine. SSNs, credit cards, credentials, and API keys are killed. Phone numbers and addresses are redacted. Fully configurable.
constraintsOptional LLM enrichment with hard daily and hourly caps. When budget is exhausted, falls back to deterministic heuristics. Never stops running, just reduces depth.
budgetOne pattern at every scale: agent-context-signal. Workers, supervisors, and control layers are the same abstraction.
| Source | Auth | Signals ingested |
|---|---|---|
| GitHub | gh auth login |
PRs, issues, commits, reviews, comments |
| Linear | LINEAR_API_KEY |
Issues, projects, cycles, comments |
| Slack | SLACK_BOT_TOKEN |
Channel messages, threads, reactions |
| Grafana | GRAFANA_API_KEY |
Dashboard metrics, Tempo traces, alerts |
Up and running in under a minute. No API keys needed for mock mode.
pip install stigmergyOr clone and install in dev mode:
git clone https://github.com/jmcentire/stigmergy.git
cd stigmergy
pip install -e ".[dev]"stigmergy initInteractive setup walks you through sources, identity, and budget configuration. Creates .stigmergy/config.yaml.
# Mock data (no keys needed)
stigmergy run --once
# Live sources
stigmergy run --once --live
# Continuous monitoring
stigmergy run --liveModular by design. Each package owns one concern.
src/stigmergy/
adapters/ Source connectors (GitHub, Linear, Slack, Grafana — mock + live)
attention/ Attention model, portfolio scoring, surfacing
cli/ CLI entry point, config, budget, live adapters
constraints/ Output filtering (PII/credential kill and redaction)
core/ Algorithms: familiarity, consensus, energy, lifecycle
delivery/ Output delivery framework
identity/ Person identity resolution across sources
mesh/ ART mesh: routing, workers, topology, insights, stability
pipeline/ Signal ingestion pipeline
policy/ Policy engine, spectral analysis, budget enforcement
primitives/ Data types: Signal, Context, Agent, Assessment
services/ LLM, embedding, vector store, token budget
structures/ Bloom filters, LSH, SimHash, ring buffers, tries
unity/ Field equations, eigenmonitor, PID control
tracing/ Execution tracingBuilt on established theory from computational neuroscience, biology, economics, and cybernetics.
Stable category formation with vigilance-gated plasticity. New patterns create new categories; familiar patterns reinforce existing ones. Solves the stability-plasticity dilemma.
Coordination through shared environment rather than direct communication. Agents leave traces in the medium; other agents respond to those traces. No central coordinator needed.
Information degradation under strategic communication. When sender bias exceeds 1/4, communication collapses to babbling equilibrium. Used to detect organizational signal decay.
Beer's cybernetic model for organizational viability. System 4 intelligence function: scanning the environment and modeling the future. Algedonic signals for pain/pleasure alerting.
Anomaly detection via Laplacian eigenvalue distribution. Structural changes in the communication graph surface as spectral perturbations before they become visible in metrics.
Stop-on-first-accept routing: the first worker above threshold takes the signal. Bounded rationality over exhaustive optimization. Fast, stable, and biologically plausible.