FintheFinder
Deep-research agent whose runs suspend until a human approves, with a deterministic citation audit that keeps AI out of the verification path.
Proof ledger
- Classification
- Built artifact
- Repository
- Alexi5000/FintheFinder
- Language
- TypeScript
- Licence
- Custom licence (GitHub reports NOASSERTION)
- Status
- Maintained
- Tests
- 3 recorded gates
- Last verified
- 2026-07-04
What it is
Most human-in-the-loop demos ask for approval after the work is done. FintheFinder suspends the run itself at the gate, so nothing downstream executes until a person approves. Citations are then audited deterministically rather than by asking a model whether its own sources were real.
Architecture
- Mastra-based run graph with suspend and resume at approval boundaries.
- Deterministic citation audit: no model participates in verifying a source.
- Contracts directory holds the typed interfaces between research, approval, and audit stages.
- Supabase for persistence, Next.js for the operator surface.
Tests and gates
- GitHub Actions ci workflow plus a tests directory and a Playwright configuration for end-to-end runs.
- Type-test tsconfig is checked in separately from the build tsconfig.
- Docker Compose brings up the full stack for reproducible local runs.
Limits
- Approval gates only help if a human actually reads the payload.
- The citation audit verifies that a source resolves and matches the claim's anchor; it does not judge source quality.
- Search coverage depends on the configured providers.
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.
AI Production Systems at TechTide AI→© 2026 Alex Cinovoj · TechTide AI · Columbus, OH