A self-hosted inbox for feedback from AI coding agents. The Go API stores multi-model review results and tooling or documentation friction reports in PostgreSQL. The included Bash skill submits, queries, and marks that feedback processed.
Use it to make feedback from separate agent sessions and machines actionable in one place. It is a telemetry store—not a review runner, benchmark, dashboard, or automated processor.
- Write-once review and friction submissions, with replay protection and friction duplicate absorption.
- Filtered retrieval and batch processed/unprocessed marking.
- API-key authentication, automatic PostgreSQL migrations, health/readiness, Prometheus metrics, and optional OpenTelemetry tracing.
- A harness-agnostic client skill with local spooling and retry on a later invocation.
The current stable source release is v1.0.1. Clone it and check out the tag for a fixed snapshot:
git clone https://github.com/foae/agent-feedback.git
cd agent-feedback
git checkout v1.0.1For a local stack, you need Git, Docker with Compose 2.24.4 or newer, Bash,
curl, and openssl. Follow the getting-started guide
for the configuration and first request. The bundled client additionally needs
jq.
- Getting started — local installation, configuration, first API request, and client skill.
- Security — trust boundary, credentials, network exposure, submitted data, and retention.
- Development — source setup, checks, tests, and code conventions.
- Deployment — image-based SSH deployment and remote stack configuration.
- Operations — probes, backups, restoration, retention, and key rotation.
- API contract — endpoint schemas, limits, errors, and idempotency.
- Architecture, conventions, patterns, and extension guide.
- Releases — versioning, publication, and upgrade notes.
The project is licensed under the MIT License.