Built on Agno

DevTools for
the agentic
era_

Every span, tool call, and model decision — visible, inspectable, and live. Built for teams shipping agents to production.

Try DevTools
Your database PostgreSQL · SQLite
AgentOS API
GET /sessionsGET /traces
Agno DevTools ● Live
Multi-agent systems
Full span hierarchy
Live from AgentOS
Tool call inspection
Memory traces
Token analytics
Zero install
Error surfacing
Session replay
P50 latency
Multi-agent systems
Full span hierarchy
Live from AgentOS
Tool call inspection
Memory traces
Token analytics
Zero install
Error surfacing
Session replay
P50 latency
01
Full span hierarchy
Every agent call, LLM invocation, tool execution, and memory update. Hierarchical, timestamped, inspectable to raw input and output.
02
Live from AgentOS
Reads directly from GET /traces and GET /sessions. One URL. No proxy, no middleware.
03
Zero install
A single HTML file. No npm, no build step, no config. Open in any browser, point at your endpoint, start debugging.
04
Mock data included
Ships with realistic mock traces — multi-agent teams, tool calls, errors. Explore every feature before wiring to a live system.
Setup

Wire it up
in minutes_

Three steps. AgentOS already generates the traces. DevTools just reads them.

1
Enable tracing

Add tracing=True to AgentOS. This writes every span to your database automatically.

2
Allow CORS

Add CORSMiddleware so the browser can reach your local server. One import, one line.

3
Open and connect

Open the HTML file. Enter your endpoint. Click Connect. Sessions load instantly.

agent_os.py
from agno.agent import Agent from agno.models.anthropic import Claude from agno.db.sqlite import SqliteDb from agno.os import AgentOS from fastapi.middleware.cors import CORSMiddleware agent = Agent( name="My Agent", model=Claude(id="claude-sonnet-4-6"), db=SqliteDb(db_file="agno.db"), ) # Enable tracing — writes spans to DB on every run agent_os = AgentOS(agents=[agent], tracing=True) app = agent_os.get_app() # Allow browser access from local HTML file app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"], )
terminal
$ uvicorn agent_os:app --reload INFO Uvicorn running on http://localhost:8000 # Open agno-devtools.html → enter localhost:8000 → Connect
Layer 01

The runtime
that sees
everything_

AgentOS runs your agents as a stateless FastAPI service. Every execution — every model call, every tool invocation — is recorded as a span and written to your database in real time.

~2μsagent instantiation
20+event types captured
100%data in your cloud
Architecture overview
Layer 02

AgentOS
exposes a clean
REST API_

Gnosis reads four endpoints. That's the entire integration. No SDK, no agent, no additional config. If AgentOS is running, Gnosis works.

GET/health Connection check
GET/sessions List all sessions
GET/traces?session_id Traces for a session
GET/traces/{id} Full span tree
session_response.json
{ "data": [ { "session_id": "ses_inv_001", "session_type": "team", "agent_name": "Investment Research Team", "created_at": "2026-04-21T10:22:10Z", "updated_at": "2026-04-21T10:22:22Z", "run_count": 6 }, // ... more sessions ], "total": 14, "page": 1, "limit": 20 }
Layer 03

Every run
is a trace.
Every trace
is a tree_

When you call agent.run(), Agno creates a RunOutput with a recursive span tree. Every decision, every branch, every retry — captured as a node in that tree.

RunOutput TraceNode store_events=True OpenTelemetry
trace_detail.json — /traces/tr_a1
{ "id": "tr_a1b2c3", "name": "Research: Solana AI Q2 2026", "type": "TEAM", "status": "OK", "duration": "12.4s", "spans": [ { "type": "AGENT", "name": "agent.run (Researcher)", "duration": "5.2s", "spans": [ { "type": "TOOL", "name": "web_search", "duration": "0.9s" }, { "type": "LLM", "name": "llm.invoke", "duration": "3.6s", "tokens": { "prompt": 3200, "completion": 5220 } } ] } ] }
Layer 04

20+ span
event types.
All surfaced_

Agno emits a distinct event for every stage of execution. Gnosis maps each type to a visual representation with color coding, timing bars, and raw data inspection.

AGENTagent.run() lifecycle — start, content, complete
TEAMMulti-agent coordination and member responses
LLMModel request + completion, tokens, latency
TOOLTool call — arguments, result, error
MEMORYMemory read/write and session summary
REASONINGChain-of-thought steps when reasoning=True
Span waterfall — 12.4s run
Layer 05

Two lines
to unlock
full visibility_

Enable tracing on AgentOS. Add CORS. That's the entire integration. Gnosis reads the rest automatically — sessions, traces, span trees, token counts, latency — all of it.

agent_os.py — complete setup
# 1. Create your agent (any model, any DB) agent = Agent( name="My Agent", model=Claude(id="claude-sonnet-4-6"), db=SqliteDb(db_file="agno.db"), ) # 2. Wrap in AgentOS — enable tracing agent_os = AgentOS( agents=[agent], tracing=True # ← this is the only new line ) app = agent_os.get_app() # 3. Allow browser access app.add_middleware( CORSMiddleware, allow_origins=["*"], ) # 4. Run — then open Gnosis and connect # uvicorn agent_os:app --reload
scroll to explore
01 / 05

What the Agno community
is building_

"Mark my words. Next big startup will be built on @AgnoAgi… and it might be mine."

@lebrxndxng

"The hype is real. @AgnoAgi is what you've been looking for. So many new toys to play with."

@smilingnosrati

"@AgnoAgi's framework is awesome. Build agents, teams, tools, workflows — it's just super flexible."

@bernhard_me

"After using Langgraph for a while, I'm starting new projects only with @AgnoAgi. Everything just makes more sense."

@IdanP70

"I'm actually very surprised how fast it is to get @AgnoAgi agents up and running. Like literally 2 minutes."

@MitchBernstein

"I don't highlight this enough: the Memory & Knowledge system in @AgnoAgi is insanely powerful."

@maxbraglia

"@AgnoAgi is the leader in agent frameworks right now. Well engineered, more intuitive, and faster."

@LamarDealMaker

"I have been using @AgnoAgi for a while now — it is so much easier to use than other frameworks. Fast too!"

@martintechlabs

"Mark my words. Next big startup will be built on @AgnoAgi… and it might be mine."

@lebrxndxng

"The hype is real. @AgnoAgi is what you've been looking for. So many new toys to play with."

@smilingnosrati

"@AgnoAgi's framework is awesome. Build agents, teams, tools, workflows — it's just super flexible."

@bernhard_me

"After using Langgraph for a while, I'm starting new projects only with @AgnoAgi. Everything just makes more sense."

@IdanP70

"I'm actually very surprised how fast it is to get @AgnoAgi agents up and running. Like literally 2 minutes."

@MitchBernstein

"I don't highlight this enough: the Memory & Knowledge system in @AgnoAgi is insanely powerful."

@maxbraglia

"@AgnoAgi is the leader in agent frameworks right now. Well engineered, more intuitive, and faster."

@LamarDealMaker

"I have been using @AgnoAgi for a while now — it is so much easier to use than other frameworks. Fast too!"

@martintechlabs
Live demo

Your agents, fully visible_

Connect to your AgentOS — or explore the mock data below

Mock data
agno-devtools · mock://demo
DEMO
Sessions5
Traces—
All Agent LLM Tool Error
Select a session
Span Detail
Select a trace
Span details appear here
5
Sessions
—
Traces
—
Tokens
0
Errors
—
P50 Latency
Documentation

Everything you need_

One HTML file. No dependencies. Works with any AgentOS instance.

Quick start Connecting Sessions Traces Span types API reference Mock data Self-hosting

Quick start

DevTools is a single HTML file. No npm, no build step, no install. Download it and open it in any browser.

# 1. Download curl -o agno-devtools.html https://agno.ai/devtools # 2. Open in browser open agno-devtools.html # 3. Start your AgentOS uvicorn agent_os:app --reload # 4. Enter localhost:8000 and click Connect
💡No AgentOS yet? The demo loads with built-in mock data so you can explore every feature immediately.

Connecting

DevTools reads directly from the Agno session and trace APIs. Your AgentOS needs two things:

1
Enable tracing

Pass tracing=True to AgentOS. This writes spans to your database on every run.

2
Allow CORS

Add CORSMiddleware so the browser can reach your local server.

from agno.os import AgentOS from fastapi.middleware.cors import CORSMiddleware agent_os = AgentOS(agents=[agent], tracing=True) app = agent_os.get_app() app.add_middleware(CORSMiddleware, allow_origins=["*"])
⚠️CORS is the most common reason Connect fails. If nothing loads, open browser DevTools → Network → check the /health request.

Sessions

A session is a persistent conversation context — one user talking to one agent over time. DevTools loads sessions from GET /sessions, sorted by most recent activity.

agentSingle agent session. One user, one agent, N turns.
teamMulti-agent team. Coordinator and specialist agents working together.
workflowStructured workflow. Sequential or parallel steps with defined transitions.

Traces

A trace is a single agent run — one call to agent.run(). Each trace has a root span and a tree of children. The detail panel has five tabs:

OverviewType, status, duration, token count, model, trace ID.
InputThe raw input passed to the agent or tool.
OutputFull response or tool result. Errors shown here too.
SpansThe full child span tree. Expand any node to drill deeper.
MetadataRaw span attributes — model name, token counts, agent IDs.

Span types

Every node in the trace tree has a type. DevTools highlights them differently in the span panel.

AGENT

A single agent run. Contains LLM calls, tool executions, and memory operations.

TEAM

A multi-agent team run. Contains multiple AGENT spans as children.

LLM

A model invocation. Includes prompt tokens, completion tokens, model name.

TOOL

A tool or function call. Arguments passed in, result returned.

API reference

DevTools uses four AgentOS endpoints — all part of the standard Agno REST API.

GET/healthConnection check on Connect click.
GET/sessions?limit=20List sessions, most recent first.
GET/traces?session_id={id}List traces for a session.
GET/traces/{trace_id}Full trace with recursive span tree.

Mock data

Ships with built-in mock data matching the real Agno schema — five sessions, fourteen traces, full span trees. Replaced automatically when you connect to a live endpoint.

💡Use mock mode to demo DevTools to your team before wiring it to a live AgentOS. All interactions work identically.

Self-hosting

A single static HTML file. Host it anywhere — S3, Vercel, Netlify, or your laptop. No server required.

# Serve locally npx serve . # Host on S3 aws s3 cp agno-devtools.html s3://your-bucket/ --acl public-read # Or just open it directly open agno-devtools.html

The future runs
on AgentOS_

Your agents are generating traces right now. DevTools makes them readable.

Try DevTools Learn about Agno