The TeamSpec reference implementation. A working executive AI assistant built to the full TeamSpec spec — showing exactly what a well-defined agent team looks like when it’s deployed and running.
Siggy is not a demo. It is a real, working implementation of the TeamSpec specification — built to show that well-defined, reliable agent teams are achievable today, not someday. If you want to understand what a TeamSpec-compatible agent looks like in practice, Siggy is the clearest possible answer.
Schedules, reschedules, and protects calendar time. Books internal meetings autonomously. Flags conflicts. Prepares agendas. Manages availability across time zones.
Researches projects, topics, customers, and prospects on demand. Synthesizes findings into structured briefings. Delegates deep-dive tasks to specialized sub-agents.
Drafts messages in the executive's voice. Sends approved communications. Manages response queues. Escalates when human judgment is required.
Monitors the pipeline for stalled opportunities. Identifies impediments blocking progress. Surfaces patterns across deals, projects, and teams with recommended actions.
Delegates tasks to both specialized AI sub-agents and human team members. Tracks completion. Follows up. Escalates overdue items. Closes the loop.
Coordinates logistics for meetings across internal and external participants. Prepares pre-read materials. Captures action items post-meeting. Ensures nothing falls through.
The TeamSpec spec covers 7 dimensions that separate well-built agents from fragile ones. Siggy implements all 7 — not as abstractions, but through specific, observable behaviors you can study and replicate.
"Prepare me for Thursday's board meeting" is not a single task. Siggy decomposes it: pull relevant board materials, research open action items, check calendar conflicts, identify attendees needing pre-briefs, synthesize everything into a structured briefing. Each sub-task is tracked independently with its own completion state.
Siggy knows what it can do without asking and what requires approval before acting. Booking internal meetings: autonomous. Sending an external commitment on behalf of the executive: requires explicit approval. Spending above a configured budget threshold: requires approval before proceeding. Budget limits are enforced as structural authority boundaries in AgentHub and checked by Forge at runtime — not left to the agent's judgment.
Data access is scoped: Siggy sees what it needs for the current task and nothing more. Message dispatch has thresholds — low-stakes internal messages proceed automatically, external communications require review. CRM data access enforces privacy guardrails that prevent exposure of sensitive deal or contact information outside approved contexts.
When a calendar API fails mid-task, Siggy detects it, queues the dependent actions, retries with exponential backoff, and notifies the user with current status — rather than silently failing or producing a partial result that looks complete. Budget exhaustion is treated the same way: Siggy detects when a run approaches its cost limit, surfaces the situation to the user, and halts cleanly rather than continuing unchecked. Every failure path — including financial ones — has a defined behavior.
Siggy communicates outcomes in plain, actionable language: "Thursday brief ready: 3 open actions, 2 calendar conflicts resolved, 1 item needs your decision — approve the vendor contract extension." No logs. No jargon. No status that requires interpretation. The human always knows exactly where things stand and what is needed.
Complex tasks don't block. Siggy delegates to specialized sub-agents — a research agent, a calendar agent, a messaging agent — and runs them in parallel via Forge pipelines. Each sub-agent operates within its own cost budget; the orchestrator tracks aggregate spend across the team in real time. If a sub-agent approaches its limit, it is paused or halted before the overall task budget is blown.
Every action Siggy takes is logged: what it decided, what tools it called, what data it accessed, what it sent, and what it deferred. Cost is part of every record — token usage and model spend are captured per action, per sub-agent, and per run. Budget vs. actual is always visible: what was allocated, what was consumed, and where the spend went. If something goes wrong — or if anyone asks what a run cost and why — the answer is always available.
FastBytes benchmarked three platforms against the Siggy implementation — scoring each on the 7 TeamSpec dimensions. The results show clearly which platforms make it easy to build reliable agent teams and which require significant extra work.
Strongest on authority management and trust/safety — the dimensions most directly shaped by building against a spec. Gaps in task decomposition flexibility and communication clarity. Best overall score among the platforms evaluated.
Strongest on task decomposition, failure handling, and scalability — reflecting a mature ecosystem with LangGraph and LangSmith. Authority management required significant custom implementation. Strong with investment; weak out of the box on compliance dimensions.
Strongest on communication clarity — bespoke builds can be tuned exactly to an individual's preferences. Critical gaps across authority management, trust/safety, failure handling, and observability. Excellent for personal use; not viable at enterprise scale without major additional investment.
The full case study details evaluation methodology, per-dimension scores, and implementation guidance for each platform.
Read the Full Case Study →The TeamSpec specification is open source. Everything Siggy demonstrates — the AgentHub config structure, the Forge execution model, all 7 dimensions — is available for you to study and build against. Whether you’re building an executive assistant, a research agent, or something entirely different, you can start from a working foundation.
The full TeamSpec specification lives on GitHub. It defines all 7 dimensions in precise, implementable terms — not abstract principles but concrete, testable requirements.
Read the Spec →The Siggy codebase is open source. Read the implementation to see how each dimension was built — the authority management config, the failure handling patterns, the observability instrumentation.
github.com/TeamSpecAI/siggy →AgentHub and Forge are the infrastructure layer that Siggy runs on. They are both open source and available for you to self-host today. Start with the tools, not just the spec.
Explore AgentHub + Forge →FastBytes.io builds and deploys TeamSpec-compatible agents for businesses and teams. If you want to ship a well-built agent team faster than building it from scratch, the FastBytes team can help.
Visit FastBytes.io →Siggy demonstrates that well-built, reliable agent teams are achievable today — for any use case, at any scale. The TeamSpec spec, AgentHub, and Forge are everything you need to go from definition to deployment. Open source. Free to use. Ready to run.