alexcinovoj.devby TechTide AI
Contact
Back to proof
Built artifact

TechTideAI2

Company-scale agent platform on a three-plane control/data/evidence architecture, with state-machine transitions, hard execution timeouts, and an adversarial eval harness over a 33-task golden suite.

Proof ledger

Classification
Built artifact
Repository
Alexi5000/TechTideAI2
Language
TypeScript
Licence
MIT
Status
Active development
Tests
3 recorded gates
Last verified
2026-08-01

What it is

TechTideAI2 separates the plane that decides from the plane that stores and the plane that proves. Agent progress is a state machine rather than a free-running loop, execution has hard timeouts, and an adversarial eval harness runs against a golden suite in CI.

Architecture

  • Three planes: control (decisions and policy), data (application state), evidence (audit and eval output).
  • Explicit state-machine transitions per agent run instead of open-ended loops.
  • Hard execution timeouts at the runtime boundary.
  • Adversarial eval harness with a 33-task golden suite, run from its own workflow.
  • Fastify and LangChain/Mastra on the service side, React front end, Supabase data plane.
Control flow
  +----------------+   +----------------+   +-----------------+
  | control plane  |   | data plane     |   | evidence plane  |
  | policy, state  |-->| app state      |   | audit + evals   |
  | machine, gates |   | supabase       |-->| golden suite    |
  +-------+--------+   +----------------+   +--------+--------+
          |                                          ^
          +------------------ traces ----------------+

Tests and gates

  • Four workflows: ci, pr, evals, and notebooks. Evals run separately from unit CI.
  • A 33-task golden suite is referenced in the README and driven by the evals workflow.
  • Contracts and database directories are versioned alongside the services.

Limits

  • The golden suite is a regression net, not a benchmark against other platforms.
  • Three-plane separation is architectural; it does not by itself satisfy any specific compliance regime.
  • Demo walkthrough content in the repository is illustrative, not a customer deployment.

Source artifacts

Patterns applied

Acceptance tests for agent work

Define done before the agent starts.

Evals that gate releases

A golden suite in CI, not a spreadsheet.

MCP permission scoping

Deny by default, declare every tool.

Observability for agent runs

One trace ID from request to receipt.

Cost controls that actually stop spend

Ceilings enforced before the call, not after the invoice.

Approval gates that suspend the run

Nothing downstream executes before a human approves.

Related proof

ClawKeeper

Governed Claude agent control plane. A deterministic policy engine runs before every LLM call, autonomous spend defaults to zero, and the audit log is enforced by the database.

TechTide Harness Kit

Trust contracts for agent skills expressed as JSON Schema: risk tiers, tool permissions, deny-by-default egress, approval gates, and prompt-injection fixtures behind a wall of CI validation gates.

CipherClaw

Deterministic multi-agent debugger with zero LLM calls in the analysis path: causal graphs, weighted root-cause scoring, and failure prediction.

Swarm 357

357 Claude agent roles across six business layers, with portable memory, enforced cost caps, a bash policy gate, and durable checkpoints.

This page is implementation proof only. Engagement scope and pricing live on TechTide AI.

Critical System Implementation at TechTide AI
Back to top
Talk to Alex
alexcinovoj.dev

© 2026 Alex Cinovoj · TechTide AI · Columbus, OH