Logfire (by the Pydantic team, MIT, ~4.4k stars) is an observability platform for AI/LLM/agent systems: natively collects LLM calls, token usage, latency and cost via OpenTelemetry, with dashboards and queries. Dual form of CLI + Python library, supporting a local preview (token-free) and a hosted platform for uploads.

In one sentence: give LLM apps a "monitoring screen" — how much each call costs, how slow it is, and which step failed, all at a glance.

You want observability for your Python app — traces, spans, metrics — without building a whole observability stack; Logfire is OpenTelemetry-based observability for Python. The two things people trip on: the first install is slow (opentelemetry dependency family, 3-5 minutes — normal, not stuck), and the "must log in?" question — local preview runs without any token (verified), only uploading to the hosted platform needs logfire auth.

When to use it:

• Tracing Python apps with minimal code (logfire.configure + instrument_*);
• Local preview of traces/spans without accounts;
• Uploading to the hosted platform with token auth.

When not to use it: you need a self-hosted full observability stack with your own retention/policies.

note: verified path here; mainland must use the Tsinghua mirror (PyPI official timeout)

# Source: Trove detailed guide · verified 2026-08-28
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple logfire==4.41.0
logfire --version

Expected output: Running Logfire 4.41.0 with CPython ...

Step 1: Confirm the library imports and configures

Why: import logfire + configure without errors means the library is installed complete and the dependency chain is fine; local preview mode passes this line without a token.

# Source: Trove detailed guide · verified 2026-08-28
python3 -c "import logfire; logfire.configure; print('logfire configured: OK')"

Expected output: logfire configured: OK (local preview runs without a token)

Stuck? How to recover:

• import fails → confirm pip and python3 share the same environment (activate the venv first);
• token error → local preview does not need a token; skip auth.

Step 2: Confirm the CLI and optional connectivity surface

Why: logfire --version confirms the CLI; when uploading to the hosted platform, logfire auth then checking ~/.logfire/ for a token confirms the auth path.

# Source: Trove detailed guide · verified 2026-08-28
logfire --version

Expected output: Running Logfire 4.41.0 with CPython ...

Stuck? How to recover:

• auth interactive stuck → use --non-interactive or the device-code flow;
• preview port taken → see global options in --help (--base-url/--region etc.).

pip installs logfire==4.41.0 (Python >= 3.10; opentelemetry family deps, first install 3-5 minutes — normal). logfire.configure is the instrumentation entry; local preview mode runs without a token; logfire auth stores the token in ~/.logfire/ for uploading telemetry.

Key concepts:

• configure: instrumentation entry point;
• preview: local preview of traces/spans (no token);
• auth: browser/device-code login (token stored in ~/.logfire/).

Where it writes: ~/.logfire/ (own directory, no shared registration surface).

• ~/.logfire/: dedicated config/token (generated, measured)
• Inside the codebase: OpenTelemetry instrumentation (logfire.instrument_* calls, code-level integration)
• Runtime: CLI commands (preview/auth), no resident daemon

Step 1: Install the package (why pin the version: 4.41.0 is the verified collection version)

Why: logfire depends on the opentelemetry family; the first install takes 3-5 minutes — be patient, it is not stuck. Mainland PyPI official times out, so the Tsinghua mirror prefix is required.

# Source: Trove detailed guide · verified 2026-08-28
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple logfire==4.41.0
logfire --version

Expected output: Running Logfire 4.41.0 with CPython ...

# Source: Trove detailed guide · verified 2026-08-28
logfire --help

Expected output: full CLI docs

After install, confirm:

• --version prints 4.41.0;
• logfire --help has full CLI docs.

Stuck? How to recover:

• install takes long → opentelemetry deps are many, normal (first 3-5 minutes); confirm the Tsinghua mirror prefix;
• Python version → confirm >= 3.10.

Step 2: Code instrumentation smoke (local preview, no token)

Why: logfire.configure is the instrumentation entry; local preview mode runs without a token — verify the library works first, then decide whether to connect the hosted platform.

# Source: Trove detailed guide · verified 2026-08-28
python3 -c "import logfire; logfire.configure; print('logfire configured: OK')"

Expected output: logfire configured: OK (local preview runs without a token)

After install, confirm:

• prints logfire configured: OK, no auth error.

Stuck? How to recover:

• token error → local preview does not need a token; if it prompts to log in, use logfire preview or skip auth;
• import fails → confirm pip and python3 share the same environment.

Step 3: Login to the hosted platform + local preview (optional)

Why: logfire auth goes through browser/device-code login (token stored in ~/.logfire/), needed only for uploading telemetry; logfire preview starts a local preview to view traces/spans.

# Source: Trove detailed guide · verified 2026-08-28
logfire auth

Expected output: browser/device-code login flow (interactive; token stored in ~/.logfire/)

# Source: Trove detailed guide · verified 2026-08-28
logfire preview

Expected output: local preview address (view traces/spans; Ctrl+C to exit)

After install, confirm:

• after auth, ~/.logfire/ has a token;
• preview opens the preview address.

Stuck? How to recover:

• auth interactive stuck → use --non-interactive or the device-code flow;
• preview port taken → see global options in --help (--base-url/--region etc.).

Startup and entry

• Instrumentation: logfire.configure in code + instrument_* calls;
• CLI: logfire --help (auth/preview etc.).

Common operations

Operation 1: Code instrumentation smoke (local preview, no token; verified here)

# Source: Trove detailed guide · verified 2026-08-28
python3 -c "import logfire; logfire.configure; print('logfire configured: OK')"

Expected output: logfire configured: OK (local preview runs without a token)

Operation 2: Login to the hosted platform (only needed for uploading)

# Source: Trove detailed guide · verified 2026-08-28
logfire auth

Expected output: browser/device-code login flow (interactive; token stored in ~/.logfire/)

Operation 3: Local preview

# Source: Trove detailed guide · verified 2026-08-28
logfire preview

Expected output: local preview address (view traces/spans; Ctrl+C to exit)

Data and directories

• Token/config: ~/.logfire/;
• In-code: logfire.instrument_* / logfire.configure calls (project code, not auto-removed).

Same source as the tool-page matrix — 19 pairs total (5 measured compatible / 14 inferred possibly compatible / 0 inferred possibly conflicting); full pairs, states and evidence: Conflict pairs (co-install matrix).

• No standalone handbook yet; for install/uninstall issues first see the agent-executable guide notes

# Source: Trove detailed guide · verified 2026-08-28
pip uninstall -y logfire logfire-api 2>/dev/null || true; rm -rf ~/.logfire

note: pip uninstall + remove ~/.logfire; in-code instrumentation is project code, not auto-deleted

# Source: Trove detailed guide · verified 2026-08-28
pip uninstall -y logfire logfire-api 2>/dev/null || true
rm -rf ~/.logfire
command -v logfire >/dev/null 2>&1 && echo "残留: logfire 仍存在" || echo "logfire removed: OK"

Expected output: logfire removed: OK

Manual checklist (code changes, not auto-deleted):

• logfire.instrument_* / logfire.configure calls in the codebase (OpenTelemetry instrumentation code)
• logfire dependency in pyproject/requirements (pip uninstall handled the local environment; project files need manual removal)

• Runtime preflight (needs Python >= 3.10)
• Official pip path (measured here); China installs must use the Tsinghua mirror (official PyPI times out); depends on the full opentelemetry family, first install 3-5 minutes
• For instrumentation-API only, install logfire-api (lightweight); uploading to the hosted platform needs logfire auth login (token in ~/.logfire/); local preview mode works token-free
• Each operation = one command block + expected output
• Heavy deps: full opentelemetry family, first install 3-5 minutes; China pip must use the Tsinghua mirror

• Q: Do I have to log in? A: No. Local preview mode runs without a token; only uploading to the hosted platform needs logfire auth.
• Q: Install very slow? A: opentelemetry deps are many, normal (first 3-5 minutes); use the Tsinghua mirror prefix domestically.
• Q: Will telemetry data leave my machine? A: only when uploading to the hosted platform (token auth); no login/local preview = no egress. Watch data compliance in production.
• Q: Conflicts with other plugins? A: judged all possible-compatible — logfire only writes ~/.logfire/ (own), no shared registration surface.
• Q: Where to report problems? A: install/uninstall issues on this marketplace go through the entry feedback channel (usage-report template); project issues go to the author's repository.

• Version: [4.41.0, 4.41.0] locked (verified collection version).
• Environment: Python >= 3.10; opentelemetry family deps (first install 3-5 minutes).
• Network: PyPI official times out, must use the Tsinghua mirror.
• Write surface: ~/.logfire/ (own, no shared registration surface).
• Data egress: only when uploading to the hosted platform (token auth); no login/local preview = no egress.
• Co-install: judged all possible-compatible — logfire only writes ~/.logfire/ (own), no shared registration surface.

• Verified: local repro log 2026-08-29 (2026-08-29, install + version + instrumentation smoke).
• Not tested: hosted platform upload (needs token); local preview long-run (per project needs).

Third-party tutorials are not individually verified; check against this page and official docs.

• 官方 README：安装与认证（pip install / logfire auth / 首次运行行为） [官方文档 · logfire 官方 · 2026-08-30（v4.41.0 README）]
  `pip install logfire` → `logfire auth` 认证（浏览器 OAuth；或设 `LOGFIRE_TOKEN` 环境变量走 write token）→ 代码里 `logfire.configure` 后开测
• 官方 README：手动 tracing API（configure / info / span / debug） [官方文档 · logfire 官方 · 2026-08-30（v4.41.0 README）]
  手动埋点三件套——`logfire.configure` 初始化、`logfire.info('Hello, {name}!', name=...) ` 结构化日志（{} 占位符）、`with logfire.span('...{question}...'):` 追踪块（span 内 debug/info 自动归属）；日期等类型自动格式化
• 官方 CLI：命令面（info / run / projects / read-tokens / inspect / gateway 等） [官方文档 · logfire 官方 · 2026-08-30（v4.41.0）]
  CLI 命令——auth（认证）/ clean（清空本地数据目录）/ gateway（本地 OAuth 代理，AI Gateway）/ inspect（检查已装包并推荐有用集成）/ whoami（当前用户与项目 URL）/ projects / read-tokens（项目读 token）/ prompt（生成给 LLM 的提示，面向 Logfire MCP）/ info（版本与环境）/ run（带 Logfire 插桩跑 Python 脚本）；global 选项 --non-interactive / --base-url（自托管）/ --region us|eu
• 官方 README：集成（FastAPI 等生态） [官方文档 · logfire 官方 · 2026-08-30（v4.41.0 README）]
  不手动埋点也可用集成——`logfire.instrument_fastapi(app)` 一键插桩 FastAPI（数据库连接器/http 库/logging handler 同理）；README 说支持很多流行包（文档页 integrations）