Scout - Agno

Scout is an open-source company intelligence agent. It navigates live information sources (web, Slack, Drive, wiki, CRM, MCP servers) to assemble context on demand, and it builds its own wiki and CRM as it learns about your company. The code is public at agno-agi/scout. The default move when working with knowledge sources is to ingest everything into a vector database, chunk, embed, and pray. Coding agents figured out the right approach. They navigate: ls, grep, open the file, follow the import. Scout does the same thing across Slack, Drive, and the rest. It searches the channel, opens the doc, expands the thread, and assembles context at question time from the live source.

How it works

Scout is a single agent with multiple context providers. Each provider exposes two natural-language tools: query_<source> for reads and update_<source> for writes, where the source supports them. This thin layer solves three problems that hit any agent with a diverse tool surface: context pollution from too many tools, degrading performance from overlapping scopes, and the main agent forgetting its job because its context is all tool quirks. A sub-agent behind each provider owns the source’s quirks. Scout sees query_slack. Behind it, a sub-agent knows to paginate by cursor and prefer conversations.replies for threads. Scout’s context never sees any of that.

Provider	Active when	Tools
Web	Always on	`query_web`. Uses the Parallel SDK when `PARALLEL_API_KEY` is set, otherwise Parallel’s free MCP server.
Workspace	Always on	`query_workspace`. Rooted at the Scout repo, so Scout can answer questions about its own codebase.
CRM	Always on	`query_crm`, `update_crm`. Contacts, projects, notes, follow-ups.
Knowledge wiki	Always on	`query_knowledge`, `update_knowledge`. Scout’s prose memory.
Voice wiki	Always on	`query_voice`. A code-managed style guide for emails, Slack, X, and long-form writing.
Slack	`SLACK_BOT_TOKEN`	`query_slack`. Read-only access to messages, channel history, threads, and users.
Google Drive	`GOOGLE_SERVICE_ACCOUNT_FILE`	`query_gdrive`. Read-only access to files, folders, and contents.
MCP	Registered in `scout/contexts.py`	One `query_mcp_<slug>` per server.

Setup for each provider is covered in the Scout README’s Context Providers section.

Self-building knowledge

Most information Scout learns from working with you is perfect for a wiki and a CRM, so it maintains both.

System	Purpose
Wiki	Scout’s prose memory: pages about your company, projects, and runbooks
CRM	People and relationships: contacts, projects, notes, follow-ups

Both start empty and grow with use. Mention that Josh from Anthropic shared a new RLM paper, and Scout adds Josh to the CRM, parses the paper into the wiki, and links them. See How Scout works in the README for how both systems work.

Run locally

You need Docker Desktop installed and running.

git clone https://github.com/agno-agi/scout && cd scout

cp example.env .env
# set OPENAI_API_KEY in .env

docker compose up -d --build

Scout is now running at http://localhost:8000. The Scout README has the full walkthrough.

Chat with Scout

Open os.agno.com and log in.
Click Add OS, choose Local, enter http://localhost:8000, then Connect.
Try the pre-configured prompts.

Deploy to Railway

Scout runs on any cloud provider. We provide scripts for Railway. Prerequisites: Railway CLI installed and railway login run.

cp .env .env.production

Edit .env.production if any values should differ from local, like a different Slack workspace or production-only credentials. The Railway scripts read .env.production first and fall back to .env.

./scripts/railway/up.sh

The up.sh script provisions PostgreSQL and the scout service, then creates your public domain. Your first deploy will fail. That’s expected. Production endpoints require RBAC authorization by default, and without a JWT_VERIFICATION_KEY the app refuses to serve traffic. Scout’s job is to keep your company data off the public web. To get your key:

Open os.agno.com, click Add OS → Live, and enter your Railway domain.
Enable Token Based Authorization.
Paste the public key into .env.production as JWT_VERIFICATION_KEY (the full PEM block, no surrounding quotes).
Sync the env. Railway auto-deploys when values change.

./scripts/railway/env.sh       # sync .env.production to Railway
./scripts/railway/redeploy.sh  # push code updates after up.sh

For production, swap the knowledge wiki to a Git-backed repo so pages survive container restarts. The README’s Deploy to Railway section covers that, plus connecting the repo to Railway for auto-deploys on every push.

Connect to Slack

Scout is designed to live in Slack as your teammate. Each Slack thread becomes a session with its own context, so follow-ups in the same thread carry forward.

Get a public URL Slack can reach: ngrok http 8000 locally, or your Railway domain in production.
Create the Slack app from the manifest in the README’s Chat with Scout in Slack section.
Set SLACK_BOT_TOKEN and SLACK_SIGNING_SECRET in .env.
Restart Scout with docker compose up -d.

Setting SLACK_BOT_TOKEN on its own activates the read-only Slack context provider. Add SLACK_SIGNING_SECRET and the Slack interface lights up too, so Scout can reply in your workspace.

Example prompts

Try these once Scout is up. Each one routes to the provider that owns the answer.

Prompt	What Scout does
”Find the latest benchmark numbers for model X.”	`query_web`, with cited sources
”Save that as a note.”	`update_crm` inserts into `scout.scout_notes`
”File a runbook for incident response.”	`update_knowledge` writes a markdown page under `wiki/knowledge/runbooks/`
”Track my coffee consumption: flat white, extra shot.”	`update_crm` creates `scout.scout_coffee_orders` and inserts the row. Schema on demand.
”Draft a Slack message announcing the launch.”	`query_voice` loads the style guide, then Scout drafts in that voice

Run evals

Scout ships three eval tiers. PostgreSQL must be running for every tier: docker compose up -d scout-db.

Tier	Command	What it catches
Wiring	`python -m evals wiring`	Tool shape drift and missing schema guards. Code-level invariants, no LLM.
Behavioral	`python -m evals`	Wrong tool choices, missing response content, forbidden tools firing.
Judges	`python -m evals judges`	Answer quality, LLM-scored.

Run a single case with python -m evals --case <id>. See EVALS.md for the full picture.

Source

The GitHub repo has the full provider setup guides under docs/ and implementation notes in AGENTS.md.

​How it works

​Self-building knowledge

​Run locally

​Chat with Scout

​Deploy to Railway

​Connect to Slack

​Example prompts

​Run evals

​Source