Agno Provider for Pragmatiks
Deploys Agno AI agents and teams to Kubernetes with reactive dependency management. Define models, tools, knowledge bases, and memory as declarative resources -- the platform handles wiring, deployment, and change propagation automatically.
Architecture
The Agno provider follows a spec pattern: most resources are stateless configuration wrappers that produce serializable specifications. Only the runner resource creates actual infrastructure.
models/openai ──┐
models/anthropic┤
├─→ agent ──┐
tools/mcp ──────┤ ├─→ runner ──→ Kubernetes Deployment + Service
tools/websearch─┤ │
knowledge ──────┤ team ─┘
vectordb/qdrant │
knowledge/content │
knowledge/embedder │
memory/manager ─┤ │
db/postgres ────┤ │
prompt ─────────┘ │
│
config (kubernetes/config) ─┘
namespace (kubernetes) ─────┘
How it works:
- Configuration resources (models, tools, knowledge, etc.) resolve their dependencies and produce a serializable
Spec - The
agentorteamresource aggregates all specs from its dependencies into a singleAgentSpecorTeamSpec - The
runnerresource deploys one or more agents and teams to Kubernetes by passing the combined specs as a JSON environment variable - The container image reconstructs every agent/team from the spec payload at startup using
from_spec()factory methods and registers them with a single AgentOS instance - When any dependency changes (e.g., a model API key rotates), Pragma propagates the change through the dependency graph and redeploys automatically
Prerequisites
- A
kubernetes/configresource providing cluster access (in-cluster, a GKE cluster, or an external kubeconfig) - A Kubernetes namespace managed by the
kubernetesprovider - API keys for your chosen model provider (OpenAI, Anthropic)
- An Agno runner container image (default:
ghcr.io/pragmatiks/agno-runner:v2) - For knowledge/RAG: a Qdrant vector database instance and an applied
qdrant/collectionsized for the embedder;knowledge/contentwrites into that collection and never creates it - For memory/sessions: a PostgreSQL database instance
Installation
pragma providers install agno
Resources
| Resource | Type Slug | Description |
|---|---|---|
| Agent | agent |
AI agent definition with model, tools, knowledge, and memory |
| Team | team |
Coordinated group of agents with shared resources |
| Runner | runner |
Deploys one or more agents/teams to Kubernetes as a Deployment + Service |
| Prompt | prompt |
Reusable instruction template with variable interpolation |
| OpenAI Model | models/openai |
OpenAI model configuration (GPT-4o, etc.) |
| Anthropic Model | models/anthropic |
Anthropic model configuration (Claude, etc.) |
| MCP Tools | tools/mcp |
Model Context Protocol server integration (stdio, SSE, streamable-http) |
| Web Search Tools | tools/websearch |
Web and news search toolkit (DuckDuckGo, Google, Bing, etc.) |
| Knowledge | knowledge |
Semantic search configuration backed by a vector database |
| Content | knowledge/content |
Content source for ingestion into a knowledge base: inline text, or a URL of a plain text, Markdown or JSON file (path ending in .txt, .text, .md, .markdown or .json). Websites, PDF, Office, CSV and YouTube URLs are refused |
| OpenAI Embedder | knowledge/embedder/openai |
OpenAI embedding model configuration |
| Qdrant VectorDB | vectordb/qdrant |
Qdrant vector database adapter for Agno knowledge. Vector search only: search_type accepts vector, because a qdrant/collection holds one unnamed dense vector and keyword or hybrid search needs sparse vectors |
| Memory Manager | memory/manager |
Agent memory management with PostgreSQL storage |
| PostgreSQL DB | db/postgres |
PostgreSQL database connection for sessions, memory, and storage |
Example: Full Agent Deployment
A realistic example showing a model, tools, knowledge base, and agent deployed to Kubernetes.
# 1. Model
provider: agno
resource: models/anthropic
name: claude
config:
id: claude-sonnet-4-20250514
api_key:
ref: secrets/anthropic-key
field: value
---
# 2. MCP tool server
provider: agno
resource: tools/mcp
name: search-tool
config:
url: http://mcp-search.tools.svc.cluster.local/sse
transport: sse
---
# 3. Web search tool
provider: agno
resource: tools/websearch
name: web-search
config:
backend: duckduckgo
enable_news: true
---
# 4. Embedder for knowledge
provider: agno
resource: knowledge/embedder/openai
name: embedder
config:
id: text-embedding-3-small
api_key:
ref: secrets/openai-key
field: value
---
# 5. Collection the content is written into, sized for text-embedding-3-small
provider: qdrant
resource: collection
name: docs
config:
url:
ref: qdrant/database/main
field: url
api_key:
ref: qdrant/database/main
field: api_key
name: docs
vectors:
size: 1536
distance: Cosine
---
# 6. Vector database adapter
provider: agno
resource: vectordb/qdrant
name: doc-vectors
config:
url:
ref: qdrant/database/main
field: url
collection:
ref: qdrant/collection/docs
field: name
api_key:
ref: qdrant/database/main
field: api_key
search_type: vector
embedder:
ref: agno/knowledge/embedder/openai/embedder
---
# 7. Knowledge base
provider: agno
resource: knowledge
name: docs-kb
config:
vector_db:
ref: agno/vectordb/qdrant/doc-vectors
max_results: 5
---
# 8. Content ingestion
provider: agno
resource: knowledge/content
name: product-docs
config:
knowledge:
ref: agno/knowledge/docs-kb
url: https://docs.example.com/product.md
description: Product documentation
---
# 9. Database for sessions and memory
provider: agno
resource: db/postgres
name: agent-db
config:
connection_url:
ref: cloudsql/instance/main
field: connection_url
username:
ref: secrets/db-creds
field: username
password:
ref: secrets/db-creds
field: password
---
# 10. Memory manager
provider: agno
resource: memory/manager
name: agent-memory
config:
db:
ref: agno/db/postgres/agent-db
add_memories: true
update_memories: true
---
# 11. Prompt template
provider: agno
resource: prompt
name: system-prompt
config:
template: |
You are {{role}}, a helpful assistant for {{company}}.
Always be concise and accurate.
variables:
role: Senior Support Engineer
company: Acme Corp
---
# 12. Agent definition
provider: agno
resource: agent
name: support-agent
config:
model:
ref: agno/models/anthropic/claude
tools:
- ref: agno/tools/mcp/search-tool
- ref: agno/tools/websearch/web-search
knowledge:
ref: agno/knowledge/docs-kb
memory:
ref: agno/memory/manager/agent-memory
db:
ref: agno/db/postgres/agent-db
prompt:
ref: agno/prompt/system-prompt
markdown: true
enable_agentic_memory: true
---
# 13. Deploy to Kubernetes
provider: agno
resource: runner
name: support-agent
config:
agents:
- ref: agno/agent/support-agent
config:
ref: kubernetes/config/main-cluster
namespace:
ref: kubernetes/namespace/agents
replicas: 2
cpu: 500m
memory: 2Gi
security_key:
ref: secrets/runner-security-key
field: value
Common Patterns
Team of Agents
Compose multiple specialized agents into a coordinated team.
provider: agno
resource: agent
name: researcher
config:
model:
ref: agno/models/openai/gpt4o
tools:
- ref: agno/tools/websearch/search
instructions:
- "You are a research specialist. Find and summarize information."
---
provider: agno
resource: agent
name: writer
config:
model:
ref: agno/models/anthropic/claude
instructions:
- "You are a technical writer. Create clear, structured content."
---
provider: agno
resource: team
name: content-team
config:
members:
- ref: agno/agent/researcher
- ref: agno/agent/writer
model:
ref: agno/models/anthropic/claude
delegate_to_all_members: true
markdown: true
---
provider: agno
resource: runner
name: content-team
config:
teams:
- ref: agno/team/content-team
config:
ref: kubernetes/config/main-cluster
namespace:
ref: kubernetes/namespace/agents
Knowledge-Augmented Agent (RAG)
An agent with access to a vector knowledge base for semantic search.
provider: qdrant
resource: collection
name: knowledge
config:
url: http://qdrant.databases.svc.cluster.local:6333
name: knowledge
vectors:
size: 1536
---
provider: agno
resource: vectordb/qdrant
name: kb-vectors
config:
url: http://qdrant.databases.svc.cluster.local:6333
collection:
ref: qdrant/collection/knowledge
field: name
search_type: vector
---
provider: agno
resource: knowledge
name: product-kb
config:
vector_db:
ref: agno/vectordb/qdrant/kb-vectors
max_results: 10
---
provider: agno
resource: knowledge/content
name: faq
config:
knowledge:
ref: agno/knowledge/product-kb
text_content: |
Q: What are the supported regions?
A: US-East, EU-West, and APAC.
---
provider: agno
resource: agent
name: support-bot
config:
model:
ref: agno/models/openai/gpt4o
knowledge:
ref: agno/knowledge/product-kb
instructions:
- "Answer questions using the knowledge base. Cite sources."
Multi-Tool Agent
An agent with multiple tool integrations.
provider: agno
resource: tools/mcp
name: github-mcp
config:
command: npx -y @modelcontextprotocol/server-github
env:
GITHUB_TOKEN:
ref: secrets/github-token
field: value
---
provider: agno
resource: tools/mcp
name: slack-mcp
config:
url: http://mcp-slack.tools.svc.cluster.local/sse
transport: sse
headers:
Authorization: "Bearer my-token"
include_tools:
- send_message
- read_channel
---
provider: agno
resource: tools/websearch
name: web
config:
backend: auto
enable_news: true
---
provider: agno
resource: agent
name: ops-agent
config:
model:
ref: agno/models/anthropic/claude
tools:
- ref: agno/tools/mcp/github-mcp
- ref: agno/tools/mcp/slack-mcp
- ref: agno/tools/websearch/web
instructions:
- "You are an operations assistant with access to GitHub, Slack, and web search."
Metadata
Release files for pragmatiks-agno-provider 2.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| pragmatiks_agno_provider-2.1.0.tar.gz | 41.9 kB | Without provenance | Details for pragmatiks_agno_provider-2.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| pragmatiks_agno_provider-2.1.0-py3-none-any.whl | Python 3 | none | any | Details for pragmatiks_agno_provider-2.1.0-py3-none-any.whl |
Total release size: 99.0 kB
Release files / pragmatiks_agno_provider-2.1.0.tar.gz
| Download URL | pragmatiks_agno_provider-2.1.0.tar.gz |
|---|---|
| Size | 41.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
6c909a87477d9c97a6f0145646ab6daa6c5a580513c3518d490bdf28eab165d4
|
|
BLAKE2b-256 checksum How to use checksums |
1bc4733f3abfb0fd561b4d96d90d4a2c0e2e9e818e7606bc64846ffe87773a03
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Release files / pragmatiks_agno_provider-2.1.0-py3-none-any.whl
| Download URL | pragmatiks_agno_provider-2.1.0-py3-none-any.whl |
|---|---|
| Size | 57.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
075d9069c81c1158e3ef88cfce25dab53861a359b88fcc35061d523f68532489
|
|
BLAKE2b-256 checksum How to use checksums |
c6ddfb2555567e934554df51a6b4c07498b9a0c3f218c614b852362869a7d55a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Security for pragmatiks-agno-provider 2.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.