The open-source, high-performance AI Agent Observability, Tracing, and Self-Healing Platform.
Solas Trace is a developer-first, zero-config, low-latency agent observability and trace-debugging workspace. Written in Rust (Axum) for raw speed and Next.js 15 for a premium user experience, it acts as a transparent proxy and OpenTelemetry receiver to capture LLM runs, multi-agent conversations, loop steps, cost metrics, and error traces.
Instead of navigating heavy, disjointed stacks, Solas Trace packs a high-performance tracing engine, interactive sandbox replays, and LLM-as-a-Judge Root Cause Analysis (RCA) into a modern, beautifully cohesive tool.
Visualize your agent executions, tool invocation trees, and trace hierarchies dynamically.
Monitor token counts, latency distributions, aggregate costs, and error rates across all projects.
- ⚡ Zero-SDK Transparent Proxy: Intercept and record all completions and messages by changing only the
base_urlin your favorite OpenAI/Anthropic python or node clients. No wrappers, no complex setups. - 🕸️ Interactive Agent Topologies: Render complex multi-agent swarms as logical DAG (Directed Acyclic Graph) trees showing step relationships, database calls, tool calls, and model outputs.
- 🛠️ Side-by-Side Sandbox Replays: Edit prompts from failed runs inside the UI and re-simulate agent steps directly through the proxy to verify prompt fixes in real-time.
- 🧠 Self-Healing Root Cause Analysis (RCA): Click a failed span to invoke an LLM-as-a-judge that evaluates error logs, identifies failures (hallucinations, parsing bugs, rate limits), and drafts proposed code fixes.
- 📊 Enterprise Infrastructure: Ready-to-go Docker stack integrating PostgreSQL (metadata and auth), ClickHouse (millions of traces/second), Redis (queues), and MinIO (S3 artifact log storage).
graph TD
ClientOpenAI["OpenAI SDK Client"] -->|base_url Proxy| Proxy["Rust Axum Engine (:8000)"]
ClientClaude["Anthropic Claude SDK"] -->|messages Proxy| Proxy
OTLP["OTLP Exporter / SDKs"] -->|HTTP/gRPC Traces| Proxy
Proxy -->|Save Metadata / Auth| DB[("PostgreSQL (:5434)")]
Proxy -->|Log Spans & Metrics| CH[("ClickHouse (:8124)")]
Proxy -->|Queue Tasks / Rate Limits| Redis[("Redis (:6380)")]
Proxy -->|Store Payload Artifacts| MinIO[("MinIO S3 (:9092)")]
Proxy -->|Forward Proxy Calls| OpenAI_API["api.openai.com"]
Proxy -->|Forward Proxy Calls| Anthropic_API["api.anthropic.com"]
Dashboard["Next.js 15 Dashboard (:3000)"] -->|Query Traces & RCA| Proxy
Dashboard -->|Read Metadata & Auth| DB
solas-trace/
├── assets/ # Sleek corporate dark logos and screenshots
├── crates/
│ └── engine/ # Rust Axum + SQLx backend proxy engine
│ ├── src/
│ │ ├── db.rs # Multi-database pool connections (Postgres)
│ │ ├── handlers.rs # OTLP ingestion, RCA report generators, replays
│ │ ├── proxy.rs # Reverse proxy interceptors for LLMs
│ │ └── main.rs # Axum server bootstrap
│ └── Cargo.toml
├── ui/ # Next.js 15 + Tailwind CSS v4 + Prisma Dashboard
│ ├── src/
│ │ ├── app/ # App router pages (Traces, Analytics, Detectors)
│ │ ├── components/ # Premium glassmorphic cards and topology graphs
│ │ └── lib/ # Better-Auth clients & utilities
│ ├── prisma/ # Prisma client relational schema files
│ ├── package.json
│ └── tailwind.config.ts
├── infrastructure/ # Docker compose companion assets
│ ├── clickhouse/migrations/ # SQL schemas for trace and span analytics
│ ├── prometheus/ # Prometheus metrics target configurations
│ └── grafana/ # Grafana dashboards for platform tracking
├── docker-compose.yml # Full dev/prod services environment composer
├── run.ps1 # PowerShell automation script for local boot
└── Cargo.toml # Cargo workspace manifest
Ensure you have Docker Desktop installed and running. Start the supporting database and cache stack:
docker compose up -dNote
This starts PostgreSQL, ClickHouse, Redis, MinIO S3, Prometheus, and Grafana.
Copy the template variables file into a live environment file:
cp .env.example .envGenerate the client libraries and migrate user/workspace metadata tables to PostgreSQL:
cd ui
npm install
npm run db:generate
npm run db:pushRun the automated PowerShell startup script at the root:
./run.ps1- Next.js Dashboard:
http://localhost:3000(Redirects to/auth/sign-in— sign up to start!) - Rust Proxy Port:
http://localhost:8000/v1 - ClickHouse Native Ingestion:
http://localhost:8124
Simply adjust the base_url of your standard client. All trace, prompt payload, token cost, and latency data will be automatically logged to Solas Trace:
from openai import OpenAI
# Zero-SDK Trace Hook Setup
client = OpenAI(
api_key="your_openai_api_key",
base_url="http://localhost:8000/v1"
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system", "content": "You are a swarm agent node."},
{"role": "user", "content": "Fetch latest weather metrics."}
]
)
print(response.choices[0].message.content)Solas Trace is released under the MIT License.


