agentobs 1.0.8

AgentOBS — reference implementation of RFC-0001 AGENTOBS, the Observability Schema Standard for Agentic AI Systems

AgentOBS

The reference implementation of the AGENTOBS Standard.
A lightweight Python SDK that gives your AI applications a common, structured way to record, sign, redact, and export events — with zero mandatory dependencies.

AGENTOBS (RFC-0001) is the open event-schema standard for observability of agentic AI systems.

---

What is this?

AgentOBS (agentobs) is the reference implementation of RFC-0001 AGENTOBS — the open event-schema standard for observability of agentic AI systems.

AGENTOBS defines a structured, typed event envelope that every LLM-adjacent instrumentation tool can emit and every observability backend can consume. It covers the full lifecycle: event envelopes, agent span hierarchies, token and cost models, HMAC audit chains, PII redaction, OTLP-compatible export, and schema governance.

Think of AgentOBS as a universal receipt format for your AI application. Every time your app calls a language model, makes a decision, redacts private data, or checks a guardrail — this library gives that action a consistent, structured record that any tool in your stack can read.

---

Why use it?

Without a shared schema, every team invents their own log format. With agentobs (and the AGENTOBS standard it implements), your logs, dashboards, compliance reports, and monitoring tools all speak the same language — automatically.

Without AgentOBS	With AgentOBS
Each service logs events differently	Every event follows the same structure
Hard to audit who saw what data	Built-in HMAC signing creates a tamper-proof audit trail
PII scattered across logs	First-class PII redaction before data leaves your app
Vendor-specific observability	OpenTelemetry-compatible — works with any monitoring stack
No way to check compatibility	CLI + programmatic compliance checks in CI
Complex integration glue	Zero required dependencies — just pip install

---

Install

pip install agentobs

import agentobs  # distribution name is agentobs, import name is agentobs

Requires Python 3.9 or later. No other packages are required for core usage.

Note: The PyPI distribution is named agentobs. The Python import name remains agentobs.

Optional extras

pip install "agentobs[jsonschema]"   # strict JSON Schema validation
pip install "agentobs[openai]"       # OpenAI auto-instrumentation (patch/unpatch)
pip install "agentobs[http]"         # Webhook + OTLP export
pip install "agentobs[pydantic]"     # Pydantic v2 model layer
pip install "agentobs[otel]"         # OpenTelemetry SDK integration
pip install "agentobs[kafka]"        # EventStream.from_kafka() via kafka-python
pip install "agentobs[langchain]"    # LangChain callback handler
pip install "agentobs[llamaindex]"   # LlamaIndex event handler
pip install "agentobs[crewai]"       # CrewAI callback handler
pip install "agentobs[datadog]"      # Datadog APM + metrics exporter
pip install "agentobs[all]"          # everything above

---

Five-minute tour

1 — Trace an LLM call with the span API

import agentobs

agentobs.configure(exporter="console", service_name="my-agent")

with agentobs.span("call-llm") as span:
    span.set_model(model="gpt-4o", system="openai")
    result = call_llm(prompt)                          # your LLM call here
    span.set_token_usage(input=512, output=128, total=640)
    span.set_status("ok")

The context manager automatically records start/end times, parent-child span relationships, and emits a structured event when it exits.

---

1c — Use the high-level Trace API (new in 2.0)

import agentobs

agentobs.configure(exporter="console", service_name="my-agent")

with agentobs.start_trace("research-agent") as trace:
    with trace.llm_call("gpt-4o", temperature=0.7) as span:
        result = call_llm(prompt)
        span.set_token_usage(input=512, output=200, total=712)
        span.set_status("ok")
        span.add_event("tool_selected", {"name": "web_search"})

    with trace.tool_call("web_search") as span:
        output = run_search(query)
        span.set_status("ok")

# Inspect the trace in the terminal
trace.print_tree()
# ─ Agent Run: research-agent  [1.2s]
#  ├─ LLM Call: gpt-4o  [0.8s]  in=512 out=200 tokens  $0.0034
#  └─ Tool Call: web_search  [0.4s]  ok

print(trace.summary())
# {'trace_id': '...', 'agent_name': 'research-agent', 'span_count': 3, ...}

The Trace object works with async with too:

async with agentobs.start_trace("async-agent") as trace:
    async with trace.llm_call("gpt-4o") as span:
        response = await async_call_llm(prompt)
        span.set_status("ok")

---

1b — Auto-instrument the OpenAI client (zero boilerplate)

from agentobs.integrations import openai as openai_integration
import openai, agentobs

# One-time setup: patch the OpenAI SDK
openai_integration.patch()

agentobs.configure(exporter="console", service_name="my-agent")

client = openai.OpenAI()

with agentobs.tracer.span("chat-gpt4o") as span:
    resp = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Hello"}],
    )
    # span.token_usage, span.cost, and span.model are now populated automatically

patch() wraps every client.chat.completions.create() call (sync and async) so that token_usage, cost, and model are auto-populated on the active span from the API response — no per-call boilerplate required.

# Restore original behaviour when you're done
openai_integration.unpatch()

---

2 — Record a raw event

from agentobs import Event, EventType, Tags

event = Event(
    event_type=EventType.TRACE_SPAN_COMPLETED,
    source="my-app@1.0.0",          # who emitted this
    org_id="org_acme",              # your organisation
    payload={
        "model": "gpt-4o",
        "prompt_tokens": 512,
        "completion_tokens": 128,
        "latency_ms": 340.5,
    },
    tags=Tags(env="production"),
)

event.validate()         # raises if structure is invalid
print(event.to_json())   # compact JSON string, ready to store or ship

Every event gets a ULID (a time-sortable unique ID) automatically — no need to generate one yourself.

---

3 — Redact private information before logging

from agentobs import Event, EventType
from agentobs.redact import Redactable, RedactionPolicy, Sensitivity

policy = RedactionPolicy(min_sensitivity=Sensitivity.PII, redacted_by="policy:gdpr-v1")

# Wrap any string that might contain PII
event = Event(
    event_type=EventType.TRACE_SPAN_COMPLETED,
    source="my-app@1.0.0",
    payload={"prompt": Redactable("Call me at 555-867-5309", Sensitivity.PII)},
)
result = policy.apply(event)
# result.event.payload["prompt"] -> "[REDACTED by policy:gdpr-v1]"

Redactable is a string wrapper. You mark fields as sensitive at the point where they are created; the policy decides what to remove before the event is written to any log.

Tip — auto-redact every span: pass redaction_policy=policy to agentobs.configure() and the policy runs automatically inside _dispatch() before any exporter sees the event.

---

4 — Sign events for tamper-proof audit trails

from agentobs.signing import sign, verify_chain, AuditStream

# Sign a single event
signed = sign(event, org_secret="my-org-secret")

# Or build a chain — every event references the one before it,
# so any gap or modification is immediately detectable.
stream = AuditStream(org_secret="my-org-secret")
for e in events:
    stream.append(e)

result = verify_chain(stream.events, org_secret="my-org-secret")

This is the same principle used in certificate chains and blockchain — each event's signature covers the previous event's signature, so you cannot alter history without breaking the chain.

Tip — auto-sign every span: pass signing_key="your-secret" to agentobs.configure() and every emitted span is signed and chained automatically, with no per-event boilerplate.

---

5 — Export to anywhere

from agentobs.stream import EventStream
from agentobs.export.jsonl import JSONLExporter
from agentobs.export.webhook import WebhookExporter
from agentobs.export.otlp import OTLPExporter
from agentobs.export.datadog import DatadogExporter
from agentobs.export.grafana import GrafanaLokiExporter

stream = EventStream(events)

# Write everything to a local file
await stream.drain(JSONLExporter("events.jsonl"))

# Ship to your OpenTelemetry collector
await stream.drain(OTLPExporter("http://otel-collector:4318/v1/traces"))

# Send to Datadog APM (traces + metrics)
await stream.drain(DatadogExporter(
    service="my-app",
    env="production",
    agent_url="http://dd-agent:8126",
    api_key="your-dd-api-key",
))

# Push to Grafana Loki
await stream.drain(GrafanaLokiExporter(
    url="http://loki:3100",
    labels={"app": "my-app", "env": "production"},
))

# Fan-out: guard-blocked events -> Slack webhook
await stream.route(
    WebhookExporter("https://hooks.slack.com/your-webhook"),
  predicate=lambda e: e.event_type == "llm.guard.output.blocked",
)

Kafka source

from agentobs.stream import EventStream

# Drain a Kafka topic directly into an EventStream
stream = EventStream.from_kafka(
    topic="llm-events",
    bootstrap_servers="kafka:9092",
    group_id="analytics",
    max_messages=5000,
)
await stream.drain(exporter)

---

6 — Sync exporters for non-async workflows

from agentobs.exporters.jsonl import SyncJSONLExporter
from agentobs.exporters.console import SyncConsoleExporter

# Log all events to a JSONL file synchronously
exporter = SyncJSONLExporter("events.jsonl")
exporter.export(event)
exporter.close()

# Pretty-print events to the terminal during development
console = SyncConsoleExporter()
console.export(event)

---

7b — Register lifecycle hooks (new in 2.0)

import agentobs

@agentobs.hooks.on_llm_call
def log_llm(span):
    print(f"LLM called: {span.model}  temp={span.temperature}")

@agentobs.hooks.on_tool_call
def log_tool(span):
    print(f"Tool called: {span.name}")

# Hooks fire automatically for every span of the matching type

---

7c — Aggregate metrics from a trace file (new in 2.0)

import agentobs
from agentobs.stream import EventStream

events = list(EventStream.from_file("events.jsonl"))
summary = agentobs.metrics.aggregate(events)

print(f"Traces:  {summary.trace_count}")
print(f"Success: {summary.agent_success_rate:.0%}")
print(f"p95 LLM: {summary.llm_latency_ms.p95:.0f} ms")
print(f"Cost:    ${summary.total_cost_usd:.4f}")

---

7d — Visualize a Gantt timeline (new in 2.0)

from agentobs.debug import visualize

html = visualize(trace.spans, path="trace.html")
# Opens trace.html in a browser — self-contained, no external deps

---

8a — Semantic cache — skip redundant LLM calls

from agentobs.cache import SemanticCache, InMemoryBackend

cache = SemanticCache(
    backend=InMemoryBackend(max_size=1024),
    similarity_threshold=0.92,   # cosine similarity cutoff
    ttl_seconds=3600,
    namespace="responses",
    emit_events=True,            # emits llm.cache.hit/miss/written events
)

# Or use the @cached decorator on any async function
from agentobs.cache import cached

@cached(threshold=0.92, ttl=3600, emit_events=True)
async def call_llm(prompt: str) -> str:
    # ... real LLM call only on cache miss
    return response

reply = await call_llm("Summarise the AGENTOBS RFC in one sentence.")
# Second call with a semantically identical prompt → instant cache hit, zero tokens spent
reply2 = await call_llm("Give me a one-sentence summary of the AGENTOBS RFC.")

---

8b — Lint your instrumentation in CI

from agentobs.lint import run_checks

source = open("myapp/pipeline.py").read()
errors = run_checks(source, filename="myapp/pipeline.py")

for err in errors:
    print(f"{err.filename}:{err.line}:{err.col}: {err.code} {err.message}")
# myapp/pipeline.py:42:12: AO002 actor_id receives a bare str; wrap with Redactable()

Or run the CLI against a whole directory:

python -m agentobs.lint myapp/
# AO001  Event() missing required field 'payload'     myapp/pipeline.py:17
# AO004  LLM call outside tracer span context         myapp/pipeline.py:53
# 2 errors in 1 file.

# Plug into flake8 / ruff automatically (entry-point registered in pyproject.toml):
flake8 myapp/

---

9 — Check compliance and inspect events from the command line

agentobs check                           # end-to-end health check (config → export → trace store)
agentobs check-compat events.json        # v2.0 compatibility checklist
agentobs validate events.jsonl           # JSON Schema validation per event
agentobs audit-chain events.jsonl        # verify HMAC signing chain integrity
agentobs inspect <EVENT_ID> events.jsonl # pretty-print a single event
agentobs stats events.jsonl              # summary: counts, tokens, cost, timestamps
agentobs list-deprecated                 # list all deprecated event types
agentobs migration-roadmap [--json]      # v2 migration roadmap
agentobs check-consumers                 # consumer registry compatibility check

CHK-1  All required fields present          (500 / 500 events)
CHK-2  Event types valid                    (500 / 500 events)
CHK-3  Source identifiers well-formed       (500 / 500 events)
CHK-5  Event IDs are valid ULIDs            (500 / 500 events)
All checks passed.

Drop any of these into your CI pipeline to catch schema drift, signing failures, or schema-breaking migrations before they reach production.

---

What is inside the box

Module	What it does	For whom
agentobs.event	The core Event envelope — the one structure all tools share	Everyone
agentobs.types	All built-in event type strings (trace, cost, cache, eval, guard…)	Everyone
agentobs.config	configure() and get_config() — global SDK configuration	Everyone
agentobs._span	Span, AgentRun, AgentStep context managers — the runtime tracing API. Uses contextvars for safe async/thread context propagation. Supports async with, span.add_event(), span.set_timeout_deadline()	App developers
agentobs._trace	Trace object and start_trace() — high-level, imperative tracing entry point; accumulates all child spans	App developers
agentobs.debug	print_tree(), summary(), visualize() — terminal tree, stats dict, and self-contained HTML Gantt timeline	App developers
agentobs.metrics	aggregate() and MetricsSummary — compute success rates, latency percentiles, token totals, and cost breakdowns from any Iterable[Event]	Data / analytics engineers
agentobs._store	TraceStore — in-memory ring buffer; get_trace(), list_tool_calls(), list_llm_calls()	Platform / tooling engineers
agentobs._hooks	HookRegistry / hooks — global span lifecycle hooks: @hooks.on_llm_call, @hooks.on_tool_call, @hooks.on_agent_start, @hooks.on_agent_end. Async variants: @hooks.on_llm_call_async, @hooks.on_tool_call_async, @hooks.on_agent_start_async, @hooks.on_agent_end_async — fired via asyncio.ensure_future().	App developers / platform
agentobs._cli	9 CLI sub-commands: check, check-compat, validate, audit-chain, inspect, stats, list-deprecated, migration-roadmap, check-consumers	DevOps / CI teams
agentobs.redact	PII detection, sensitivity levels, redaction policies	Data privacy / GDPR teams
agentobs.signing	HMAC-SHA256 event signing and tamper-evident audit chains	Security / compliance teams
agentobs.compliance	Programmatic v2.0 compatibility checks — no pytest required	Platform / DevOps teams
agentobs.export	Ship events to files (JSONL), HTTP webhooks, OTLP collectors, Datadog APM, or Grafana Loki	Infra / observability teams
agentobs.exporters	Sync exporters — SyncJSONLExporter and SyncConsoleExporter for non-async code	App developers
agentobs.stream	Fan-out router — one drain() call reaches multiple backends; Kafka source via from_kafka()	Platform engineers
agentobs.validate	JSON Schema validation against the published v2.0 schema	All teams
agentobs.consumer	Declare schema-namespace dependencies; fail fast at startup if version requirements are not met	Platform / integration teams
agentobs.governance	Policy-based event gating — block prohibited types, warn on deprecated usage, enforce custom rules	Platform / compliance teams
agentobs.deprecations	Register and surface per-event-type deprecation notices at runtime	Library maintainers
agentobs.testing	Test utilities: MockExporter, capture_events() context manager, assert_event_schema_valid(), and trace_store() isolated store context manager. Write unit tests for your AI pipeline without real exporters.	App developers / test authors
agentobs.auto	Integration auto-discovery: agentobs.auto.setup() auto-patches every installed LLM integration (OpenAI, Anthropic, Ollama, Groq, Together AI). setup() must be called explicitly; agentobs.auto.teardown() cleanly unpatches all.	App developers
agentobs.integrations	Plug-in adapters for OpenAI (auto-instrumentation via patch()), LangChain, LlamaIndex, Anthropic, Groq, Ollama, Together, and CrewAI (AgentOBSCrewAIHandler + patch()). agentobs.integrations._pricing ships a static USD/1M-token pricing table for all current OpenAI models.	App developers
agentobs.namespaces	Typed payload dataclasses for all 10 built-in event namespaces	Tool authors
agentobs.models	Optional Pydantic v2 models for teams that prefer validated schemas	API / backend teams
agentobs.trace	@trace() decorator — wraps sync/async functions, auto-emits span start/end events with timing and error capture. agentobs.export.otlp_bridge converts spans to OTLP proto dicts.	App developers
agentobs.cost	CostTracker, BudgetMonitor, @budget_alert, emit_cost_event(), cost_summary() — track and alert on token spend across a session	App developers / FinOps
agentobs.inspect	InspectorSession context manager + inspect_trace() — intercept and record tool call arguments, results, latency, and errors within a trace	Platform / debugging
agentobs.toolsmith	@tool decorator + ToolRegistry — register functions as typed tools; build_openai_schema() / build_anthropic_schema() render JSON schemas for function-calling APIs	App developers
agentobs.retry	@retry with exponential back-off, FallbackChain, CircuitBreaker, CostAwareRouter — resilient LLM provider routing with observability events at each step	App developers / SREs
agentobs.cache	SemanticCache + @cached decorator — deduplicate LLM calls via cosine-similarity matching; pluggable backends: InMemoryBackend, SQLiteBackend, RedisBackend; emits llm.cache.* events	App developers / FinOps
agentobs.lint	run_checks(source, filename) — AST-based instrumentation linter; five AO-codes (AO001–AO005); flake8 plugin; python -m agentobs.lint CLI	All teams / CI pipelines

---

Event namespaces

Every event carries a payload — a dictionary whose shape is defined by the event's namespace. The ten built-in namespaces cover everything from raw model traces to safety guardrails:

Namespace prefix	Dataclass	What it records
llm.trace.*	SpanPayload, AgentRunPayload, AgentStepPayload	Model call — tokens, latency, finish reason (frozen v2)
llm.cost.*	CostPayload	Per-call cost in USD
llm.cache.*	CachePayload	Cache hit/miss, backend, TTL
llm.eval.*	EvalScenarioPayload	Scores, labels, evaluator identity
llm.guard.*	GuardPayload	Safety classifier output, block decisions
llm.fence.*	FencePayload	Topic constraints, allow/block lists
llm.prompt.*	PromptPayload	Prompt template version, rendered text
llm.redact.*	RedactPayload	PII audit record — what was found and removed
llm.diff.*	DiffPayload	Prompt/response delta between two events
llm.template.*	TemplatePayload	Template registry metadata

from agentobs.namespaces.trace import SpanPayload
from agentobs import Event

payload = SpanPayload(
    span_name="call-llm",
    span_id="abc123",
    trace_id="def456",
    start_time_ns=1_000_000_000,
    end_time_ns=1_340_000_000,
    status="ok",
)

event = Event(
    event_type="llm.trace.span.completed",
    source="my-app@1.0.0",
    payload=payload.to_dict(),
)

---

Quality standards

• 3 032 tests (2 990 passing, 42 skipped) — unit, integration, property-based (Hypothesis), and performance benchmarks
• ≥ 92.84 % line and branch coverage — measured with pytest-cov; 90 % minimum enforced in CI
• Zero required dependencies — the entire core runs on Python's standard library alone
• Typed — full py.typed marker; works with mypy and pyright out of the box
• Frozen v2 trace schema — llm.trace.* payload fields will never break between minor releases
• async-safe context propagation — contextvars-based span stacks work correctly across asyncio tasks, thread pools, and executors
• Version 1.0.7 adds: @trace() decorator, OTLP bridge, CostTracker / BudgetMonitor, InspectorSession, ToolRegistry / @tool, @retry / FallbackChain / CircuitBreaker, SemanticCache / @cached, and agentobs.lint (AO001–AO005, flake8 plugin, CLI)
• Version 2.0.0 adds: Trace / start_trace(), async with, span.add_event(), print_tree() / summary() / visualize(), sampling controls, metrics.aggregate(), TraceStore, HookRegistry, CrewAI integration
• Version 1.0.6 adds: agentobs.testing, agentobs.auto, async lifecycle hooks, agentobs check CLI, export retry with back-off, unpatch() / is_patched() for all integrations, frozen payload dataclasses, assert_no_sunset_reached()

---

Project structure

agentobs/
├── __init__.py       <- Public API surface (start here)
├── event.py          <- The Event envelope
├── types.py          <- EventType enum  (+ SpanErrorCategory)
├── config.py         <- configure() / get_config() / AgentOBSConfig
│                        (sample_rate, always_sample_errors, include_raw_tool_io,
│                         enable_trace_store, trace_store_size)
├── _span.py          <- Span, AgentRun, AgentStep context managers
│                        (contextvars stacks, async with, add_event,
│                         record_error, set_timeout_deadline)
├── _trace.py         <- Trace class + start_trace()          [NEW in 2.0]
├── _tracer.py        <- Tracer — top-level tracing entry point
├── _stream.py        <- Internal dispatch: sample → redact → sign → export
├── _store.py         <- TraceStore ring buffer                [NEW in 2.0]
├── _hooks.py         <- HookRegistry singleton (hooks)        [NEW in 2.0]
├── _cli.py           <- CLI entry-point (9 sub-commands: check, check-compat, …)
├── trace.py          <- @trace() decorator + SpanOTLPBridge   [NEW in 1.0.7]
├── cost.py           <- CostTracker, BudgetMonitor, @budget_alert [NEW in 1.0.7]
├── inspect.py        <- InspectorSession, inspect_trace()     [NEW in 1.0.7]
├── toolsmith.py      <- @tool, ToolRegistry, build_openai_schema() [NEW in 1.0.7]
├── retry.py          <- @retry, FallbackChain, CircuitBreaker [NEW in 1.0.7]
├── cache.py          <- SemanticCache, @cached, *Backend      [NEW in 1.0.7]
├── lint/             <- run_checks(), AO001-AO005, flake8 plugin, CLI [NEW in 1.0.7]
│   ├── __init__.py
│   ├── _visitor.py
│   ├── _checks.py
│   ├── _flake8.py
│   └── __main__.py
├── testing.py        <- MockExporter, capture_events(), assert_event_schema_valid(),
│                        trace_store() — test utilities without real exporters [1.0.6]
├── auto.py           <- Integration auto-discovery; setup() / teardown()        [1.0.6]
├── debug.py          <- print_tree, summary, visualize        [NEW in 2.0]
├── metrics.py        <- aggregate(), MetricsSummary, etc.     [NEW in 2.0]
├── signing.py        <- HMAC signing & audit chains
├── redact.py         <- PII redaction
├── validate.py       <- JSON Schema validation
├── consumer.py       <- Consumer registry & schema-version compatibility
├── governance.py     <- Event governance policies
├── deprecations.py   <- Per-event-type deprecation tracking
├── compliance/       <- Compatibility checklist suite
├── export/
│   ├── jsonl.py      <- Local file export (async)
│   ├── webhook.py    <- HTTP POST export
│   ├── otlp.py       <- OpenTelemetry export
│   ├── datadog.py    <- Datadog APM traces + metrics
│   └── grafana.py    <- Grafana Loki export
├── exporters/
│   ├── jsonl.py      <- SyncJSONLExporter
│   └── console.py    <- SyncConsoleExporter
├── stream.py         <- EventStream fan-out router (+ Kafka source)
├── integrations/
│   ├── langchain.py  <- LangChain callback handler
│   ├── llamaindex.py <- LlamaIndex event handler
│   ├── openai.py     <- OpenAI tracing wrapper
│   ├── crewai.py     <- CrewAI handler + patch()              [NEW in 2.0]
│   └── ...           (anthropic, groq, ollama, together)
├── namespaces/       <- Typed payload dataclasses
│   ├── trace.py        (SpanPayload + temperature/top_p/max_tokens/error_category,
│   │                    SpanEvent, ToolCall + arguments_raw/result_raw/retry_count)
│   ├── cost.py
│   ├── cache.py
│   └── ...
├── models.py         <- Optional Pydantic v2 models
└── migrate.py        <- Schema migration helpers
examples/             <- Runnable sample scripts
├── openai_chat.py    <- OpenAI + JSONL export
├── agent_workflow.py <- Multi-step agent + console exporter
├── langchain_chain.py<- LangChain callback handler
└── secure_pipeline.py<- HMAC signing + PII redaction together

---

Development setup

git clone https://github.com/veerarag1973/agentobs.git
cd agentobs

python -m venv .venv
.venv\Scripts\activate          # Windows
# source .venv/bin/activate     # macOS / Linux

pip install -e ".[dev]"
pytest                          # run all 3 032 tests

Code quality commands

ruff check .                  # linting
ruff format .                 # auto-format
mypy agentobs                  # type checking
pytest --cov                  # tests + coverage report (>=90% required)

Build the docs locally

pip install -e ".[docs]"
cd docs
sphinx-build -b html . _build/html   # open _build/html/index.html

---

Compatibility and versioning

agentobs implements RFC-0001 AGENTOBS (Observability Schema Standard for Agentic AI Systems). The current schema version is 2.0.

This project follows Semantic Versioning:

• Patch releases (1.0.x) — bug fixes only, fully backwards-compatible
• Minor releases (1.x.0) — new features, backwards-compatible
• Major releases (x.0.0) — breaking changes, announced in advance

The llm.trace.* namespace payload schema is additionally frozen at v2: even a major release will not remove or rename fields from SpanPayload, AgentRunPayload, or AgentStepPayload.

---

Changelog

See docs/changelog.md for the full version history.

---

Contributing

Contributions are welcome! Please read the Contributing Guide first, then open an issue or pull request.

Key rules:

• All new code must maintain >= 90 % test coverage
• Follow the existing Google-style docstrings
• Run ruff and mypy before submitting

---

License

MIT — free for personal and commercial use.

---

Made with care for the AI observability community.
Docs · Quickstart · API Reference · Report a bug