# Observal AI coding agent observability: sessions and traces > Source: > Published: 2026-11-15 15:30:13+00:00 # Observal AI coding agent observability: sessions and traces A developer tells you: « the agent ran for an hour, burned a pile of tokens and shipped nothing. » There is no error code and no stack trace, only a conversation. For that case, Observal AI coding agent observability means keeping the session itself, which is the ground Observal covers: an open sour ce repository (Apache 2.0) showing about 4.3k stars on the GitHub page I read on 9 November 2026. This post approaches it through the operational decision it forces on you: what to keep from an agent session, where, and for whom. A note on method: span.observer did not deploy Observal. Everything below comes from the README, the repository documentation and the screenshots it publishes; anything I could not read is marked « unverified ». First trap: the name moved. The README now calls itself « Agentic Resource Registry » and states « Formerly Observal: same code, same maintainers, same Apache 2.0 license ». The command is still observal and old links redirect. Today the project describes itself as a self-hosted registry of resources for coding agents (skills, MCP servers, hooks, prompts, sandboxes) with a session-tracing layer on top. If you were looking for an LLM benchmark-style evaluation platform, that is not what the documentation describes (see limits below). Facts taken from the repository: Apache 2.0 licence, Python on the server and CLI (FastAPI, Typer), React for the UI, PostgreSQL 16 for the registry, ClickHouse for telemetry, Redis for the queue. The latest commit on the main branch is dated 7 November 2026 and pyproject.toml carries version 1.17.0-dev.0; I could not read a published release on the page, so I quote no stable number. The fork count (about 1.1k) comes from the same page. The mechanism is easy to state. A hook or extension in the harness (Claude Code, Codex, Cursor, Copilot, Kiro, OpenCode, Pi, Goose, Antigravity according to the README) reads the conversation transcript and sends it to the server. The CLI first writes to a durable local outbox, then posts to /api/v1/ingest/session; the server returns the highest acknowledged position, and delivery is described as « at-least-once with effectively-once storage ». observal reconcile replays what is missing from the local source. Each record becomes an event: prompt, response, tool call, result, token update, error, subagent. Per-session aggregates (tokens, models, tools, duration) feed the dashboards. The documentation is frank about a useful limit: detail depends on what the harness writes, and Observal does not intercept MCP traffic or invent fields. The next screenshot comes from the repository (docs/img, covered by the project’s Apache 2.0 licence). It shows a Pi session: duration, 17 turns, input and output tokens, cache reads, tool call count, models, then the expandable list of turns with their duration. Source : github.com The session list gives the overview: platform, input and output tokens, tools called, duration, and a « Score » column that stays empty in this demo capture. It is the only evaluation hint I read there; I do not know when or how that score is filled. Source : github.com The « Debug agent failures » guide walks through four moves, which I paraphrase: find the session (filters by harness, agent, model, period), read the events in order, inspect tool call detail, then patch delivery gaps with observal reconcile. Its pattern table is the most telling part: the same tool re-called with identical arguments points to a loop; a call with no result suggests an abrupt stop or an incomplete transcript; an expected MCP call that never shows up means the model did not pick it, or the harness did not record it. The commands in the capture below are those of the README and that guide. Per the README there are two parts: a server to host (API, web UI, databases) and a CLI on each workstation. The server needs Docker Engine 24 or later with Compose v2; the install-server.sh script downloads a Compose package, generates secret files, binds ports to loopback by default and starts ten services. For the CLI: an install.sh script (macOS, Linux), install.ps1 (Windows) or uv tool install observal-cli on Python 3.11 or later. Then: observal auth login observal doctor --patch The second command detects the harness and installs the telemetry hooks. An operator’s tip: these installs go through curl | bash. Download the script, read it, and try it on a throwaway machine before pointing real sessions at it. I did not run these commands. The project applies observability to its own stack, without OpenTelemetry. According to its self-observability page: /livez and /readyz probes (the latter checks PostgreSQL, ClickHouse and Redis), JSON logs with a request ID carried by the X-Request-ID header, and Prometheus metrics on /metrics (request counter, latency histogram, memory, CPU). A Prometheus and a Grafana ship in the Compose file. On the agent side, the Claude Code integration page tells you not to set the OTEL_* variables or CLAUDE_CODE_ENABLE_TELEMETRY: Observal does not consume OTLP telemetry and relies only on its hooks and reconciliation. If you want OpenTelemetry metrics on your agents, that is a different pipeline; see monitoring coding agents in the enterprise. The project documents no Dynatrace integration, and we assume none. Evaluation: I found no benchmark or quality test set. There are LLM-generated « Insights » reports (through LiteLLM, with the provider you choose) summarising what helps or hinders, which is usage analysis, not a correctness measure. Privacy: a session contains your prompts, command output and sometimes paths or secrets. Decide on retention (the databases page mentions retention policies) and on who may read what before deploying. Maturity: a development version, a changelog with a busy « Unreleased » section, and the Leaderboard is announced there as off by default. Expect movement. Coverage: detail varies by harness; a poor transcript gives a poor trace. Weight: ten services to watch agents is a real infrastructure project. Start with one harness and one pilot team, on an isolated server. Keep ports on loopback or behind an authenticated proxy, as the default does. Monitor the stack with /readyz and Prometheus; learn to repair gaps with observal reconcile. Pair session traces with cost tracking: see AI agent cost from OpenTelemetry token metrics. Put guardrails upstream, as in AI agent guardrails in production. For a neighbouring repository, read kagent, AI agents on Kubernetes. The README presents it as Agentic Resource Registry, « formerly Observal », with the same code base and licence; the CLI is still called observal. I did not read any automatic evaluation feature: the project records sessions and produces usage reports. A score column exists in the UI, with no explanation I could verify. No: ingestion goes through hooks and reconciliation, and the documentation advises against enabling Claude Code’s OTLP telemetry for this integration. Observal answers a concrete question: what happened in that agent session? Its strength is the kept, replayable trace; its limits are a heavy stack, little visible evaluation and a project still on the move. For a team that lets agents work on its code, it deserves an isolated trial, with a clear decision about the data it keeps. span.observer is an independent community blog, not affiliated with the Observal project, its maintainers or Dynatrace LLC. Trademarks belong to their owners. Tools change quickly: check commands and options in their official documentation. Originally published on span.observer. French version: https://span.observer/fr/observal-observabilite-agents-de-code/. ## Key Takeaways - •This post approaches it through the operational decision it forces on you: what to keep from an agent session, where, and for whom. - •This story was reported by **Dev.to** , covering developments in the**dev** space. - •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage. 📖 Continue reading the full article: [Read Full Article on Dev.to →](https://dev.to/spanobserver/observal-ai-coding-agent-observability-sessions-and-traces-56j5)