Skip to content

Commit 81d8ce9

Browse files
DidierRLopesDidierRLopes
committed
update to latest 4 blogposts
1 parent c66b4b6 commit 81d8ce9

22 files changed

+595
-0
lines changed

‎blog/2025-12-02-introducing-osor-agent.md‎

Lines changed: 127 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,127 @@
1+
---
2+
slug: introducing-osor-agent
3+
title: Introducing OSOR Agent
4+
date: 2025-12-02
5+
image: /blog/2025-12-02-introducing-osor-agent
6+
tags:
7+
- ai
8+
- open-source
9+
- openbb
10+
- finance
11+
- agents
12+
- llm
13+
description: How I quickly hot-swap my financial agent with different LLM providers - fully open source
14+
hideSidebar: true
15+
---
16+
17+
![Introducing OSOR Agent](/blog/2025-12-02-introducing-osor-agent.png)
18+
19+
How I quickly hot-swap my financial agent with different LLM providers - fully open source.
20+
21+
<!-- truncate -->
22+
23+
<div style={{borderTop: '1px solid #0088CC', margin: '1.5em 0'}} />
24+
25+
When Gemini 3.0 dropped last week, I wanted to test it immediately.
26+
27+
But I have to go to Google AI Studio, and none of my data is there...
28+
29+
And I couldn't be bothered to drop CSV files or asking Claude Code to use it via API simply.
30+
31+
However, I have OpenBB workspace - which:
32+
33+
- Has the financial data I care about
34+
- AND I can bring my own agent
35+
36+
<br />
37+
38+
Silly me...
39+
40+
So, I spent Saturday building **OSOR Agent** - a bridge between two problems:
41+
42+
1. New models drop constantly, but testing them with financial data requires jumping between platforms
43+
2. OpenBB has all the financial data infrastructure, but no easy way to experiment with different LLM backends
44+
45+
<br />
46+
47+
So I built something that solves both.
48+
49+
Before we start, here's the entire agent code fully open source: [https://github.com/DidierRLopes/openbb-open-router-agent](https://github.com/DidierRLopes/openbb-open-router-agent)
50+
51+
## What OSOR Agent Does
52+
53+
OSOR Agent combines OpenBB's data & infra layer with Open Router's model flexibility.
54+
55+
Here's what that actually means:
56+
57+
### Pick Any Model, Anytime
58+
59+
Open Router gives you access to hundreds of models - Claude, GPT-4o, Gemini, Grok, open-source models.
60+
61+
Some are literally free, like Grok 4.1 Fast (as of Nov' 23).
62+
63+
<p align="center">
64+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_1.png" alt="Open Router model selection showing free models" />
65+
</p>
66+
67+
When a new model drops, you just select it. No new account, no new interface.
68+
69+
<p align="center">
70+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_2.png" alt="Model selection dropdown in OSOR Agent" />
71+
</p>
72+
73+
### Mature AI Chatting Interface
74+
75+
Instead of copy-pasting financial data around, the agent pulls directly from OpenBB widgets. Your market data, portfolio metrics, news, whatever you've connected to your workspace, feeds directly into the LLM context.
76+
77+
You can also expand AI chat so the remainder of the dashboard disappears.
78+
79+
It even has access to data via MCP as you can see below, and shares its internal step by step reasoning!
80+
81+
<p align="center">
82+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_3.png" alt="OSOR Agent showing step by step reasoning with MCP data" />
83+
</p>
84+
85+
### Your Agentic Financial Workspace
86+
87+
The agent can generate charts, tables, and citations so you're not just reading wall-of-text responses. It understands your data structure and responds accordingly.
88+
89+
<p align="center">
90+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_4.png" alt="OSOR Agent generating charts and tables" />
91+
</p>
92+
93+
In my opinion, you shouldn't have to choose between testing the latest AI model and having access to your financial data.
94+
95+
You shouldn't have to manually move data around.
96+
97+
You shouldn't have to become an expert in multiple platforms.
98+
99+
OSOR Agent is small, but the principle is big. It treats data access like infrastructure - get it right once, then let people experiment freely on top of it.
100+
101+
## Open Source
102+
103+
The full code is MIT licensed and open source.
104+
105+
That means you can fork it, extend it, build whatever you want on top of it. Want to add RAG underneath? Build it. Want custom analysis for a specific strategy? Build it.
106+
107+
The best models are worthless without access to data that matters.
108+
109+
OSOR Agent is a step in that direction on top of a mature financial workspace, OpenBB Workspace.
110+
111+
### This is a Starting Point
112+
113+
There's a lot that can be done to improve the agent, this was a weekend project built on top of our vanilla agent examples [https://github.com/OpenBB-finance/agents-for-openbb](https://github.com/OpenBB-finance/agents-for-openbb)
114+
115+
Here's an example of the agent powered by Gemini 3 working with Top Events from Polymarket when asked about what events I should look at within the "FED" category.
116+
117+
<p align="center">
118+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_5.png" alt="OSOR Agent with Polymarket FED events analysis" />
119+
</p>
120+
121+
<p align="center">
122+
<img width="800" src="/blog/2025-12-02-introducing-osor-agent_6.png" alt="OSOR Agent detailed analysis of FED events" />
123+
</p>
124+
125+
Contributions are welcome, happy hacking! :)
126+
127+
[https://github.com/DidierRLopes/openbb-open-router-agent](https://github.com/DidierRLopes/openbb-open-router-agent)

‎blog/2025-12-09-your-ai-team-isnt-doing-ai-theyre-rebuilding-infrastructure.md‎

Lines changed: 142 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,142 @@
1+
---
2+
slug: your-ai-team-isnt-doing-ai-theyre-rebuilding-infrastructure
3+
title: Your AI team isn't doing AI. They're rebuilding infrastructure.
4+
date: 2025-12-09
5+
image: /blog/2025-12-09-your-ai-team-isnt-doing-ai-theyre-rebuilding-infrastructure
6+
tags:
7+
- ai
8+
- infrastructure
9+
- enterprise
10+
- openbb
11+
- startup
12+
- finance
13+
description: From weekend POC to production - the 24-month gap nobody talks about
14+
hideSidebar: true
15+
---
16+
17+
![Your AI team isn't doing AI. They're rebuilding infrastructure.](/blog/2025-12-09-your-ai-team-isnt-doing-ai-theyre-rebuilding-infrastructure.png)
18+
19+
From weekend POC to production: the 24-month gap nobody talks about.
20+
21+
<!-- truncate -->
22+
23+
<div style={{borderTop: '1px solid #0088CC', margin: '1.5em 0'}} />
24+
25+
You know the story. Someone on your team builds a Streamlit chatbot over a weekend. They feed it your research documents, hook it up to OpenAI, and boom – it works.
26+
27+
It actually works. You demo it to leadership. Everyone gets excited.
28+
29+
*"This is it. This is our AI system."*
30+
31+
**Except that it isn't.**
32+
33+
Six months later, you realize that the weekend POC is nothing like what you actually need. Context management doesn't work at scale. The AI hallucinates in production. It can't handle complex workflows. Data security is a nightmare. You don't have audit trails. And your best engineers? They're still in the weeds, wiring APIs, debugging context windows, building data pipelines, handling compliance requirements. They're not doing AI innovation. They're stuck doing foundational infrastructure work that every other firm is also doing.
34+
35+
And guess what?
36+
37+
Our OpenBB Copilot started exactly like that. But that was more than 2 years ago.
38+
39+
Don't believe me? Look at this demo in December 2023, where we already had an AI chatbot integrated with the workspace that was able to handle structured and unstructured data.
40+
41+
<div className="flex place-items-center justify-center items-center rounded-sm mx-auto">
42+
<iframe
43+
src="https://www.youtube.com/embed/V1rYmWWVbIY"
44+
width="800"
45+
height="400"
46+
/>
47+
</div>
48+
49+
<br />
50+
51+
The difference?
52+
53+
We didn't stop there. We spent the next years, and a couple of $M in R&D, learning what it actually takes to go from "fun experiment" to "production system that financial teams can trust".
54+
55+
We learned the hard way what a CTO actually needs when you're operating at scale: sophisticated context management, proper data governance, audit trails, security that doesn't compromise workflow, AI that's genuinely agentic and not just chatty.
56+
57+
<p align="center">
58+
<img width="800" src="/blog/2025-12-09-your-ai-team-isnt-doing-ai-theyre-rebuilding-infrastructure_1.png" alt="7 level context layers slide from training workshop" />
59+
</p>
60+
<p align="center" style={{fontSize: '0.85em', marginTop: '-0.5em'}}>Slide from our training workshop deck that shows the 7 level of context layers that the agent has to navigate, with each of these layers having sometimes hundreds to thousands of datasets.</p>
61+
62+
## The foundation problem
63+
64+
Most teams approach AI the same way: start from zero, wire together APIs, paste some prompt engineering on top, and hope it scales. This works until it doesn't. You're managing context windows, debugging agent hallucinations, building retrieval pipelines, handling data governance, integrating with proprietary systems - essentially rebuilding the same foundational plumbing that hundreds of other firms are also rebuilding in parallel.
65+
66+
It's a massive waste of engineering effort.
67+
68+
Getting from "ChatGPT chatbot" to "production AI system that actually works" takes 24+ months of serious engineering work. Most teams don't have that. More importantly, your best quants and engineers get pulled away from high-value work, actual research, trading strategy, portfolio analysis, to build foundational infrastructure.
69+
70+
That's not an investment; it's opportunity cost.
71+
72+
Every month your top engineers spend on plumbing is a month they're not generating alpha or building your competitive advantage.
73+
74+
I think there's a better way.
75+
76+
## What we're offering enterprise clients
77+
78+
For our largest enterprise clients (above a certain seat count or contract size), we're providing source-available access to OpenBB Copilot. This is our AI system - the one we've spent over two years and a couple of $M in R&D building.
79+
80+
You get the code. You get the architecture. You get two years of hard-earned lessons about what actually works in production.
81+
82+
Here's what that means in practice:
83+
84+
### 1. You skip the 24-month foundational build
85+
86+
Instead of your team spending the next two years reinventing context management, agentic patterns (e.g. RAG), output validation, reasoning, citations, structured with unstructured data, URLs, MCPs, and data governance, they get a proven system from day one.
87+
88+
They can inspect the code, understand the design decisions, and extend it for your specific use cases. That's 24 months of engineering time you just reclaimed.
89+
90+
### 2. You get a system built for real-world investment workflows
91+
92+
Copilot wasn't designed in a lab - it was built to handle actual problems: connecting to multiple data sources, maintaining context across complex reasoning chains, enforcing data entitlements, handling compliance requirements, producing outputs that fit into real decision-making.
93+
94+
The architecture treats AI as a first-class component of a workflow, not a chatbot afterthought. Features like sophisticated context management (understanding which data sources matter most), dynamic widget parameter modification (automatically adjusting queries based on your needs), step-by-step reasoning visibility (so you can audit how the AI arrived at answers) and citations (even at the document sentence level) make the difference between a toy and a tool you can actually trust.
95+
96+
### 3. You can customize and own it
97+
98+
Because you have source access, you're not locked into our roadmap or dependent on our support schedule. Your team can extend it, integrate it with proprietary systems, swap out models, add your own data sources – whatever your strategy requires.
99+
100+
You build your secret sauce on top of a solid foundation instead of building the foundation itself.
101+
102+
Damn, you can even rebrand it internally to "Not OpenBB agent".
103+
104+
### 4. You're not starting from zero on security and governance
105+
106+
Enterprise AI is risky. Data leaks, compliance violations, audit trails - this stuff matters. Copilot was built with the workspace enterprise requirements from day one: on-prem deployment options, SSO, role-based access control, SOC2 compliance, proper data entitlement enforcement. Your team isn't reinventing the security wheel; they're extending a system that was built to be secure.
107+
108+
### 5. Out-of-the-box financial environment designed for it to thrive
109+
110+
Here's what most AI companies miss: a copilot is only as good as the environment it operates in. Copilot doesn't sit in isolation. It lives inside OpenBB Workspace, a purpose-built financial analysis platform with sophisticated context management, dynamic widget systems, parameter linking across dashboards, and native integration with your proprietary data sources.
111+
112+
This means Copilot understands the structure of your analysis – which data matters, which widgets are active, what parameters your team is using.
113+
114+
It's not just a chatbot bolted onto a generic system. It's an AI agent that's designed to orchestrate across your entire analytical environment. Your team gets access to the full capability stack: visualization, data integration, workflow automation, and AI reasoning - all built to work together.
115+
116+
## Why this is the best bootstrap available
117+
118+
The core insight is simple: you can't innovate on the foundation while you're building it. Your best engineers get stuck doing plumbing. Your timeline stretches. Your cost balloons. And by the time you finish, the market has moved on.
119+
120+
With source-available access to Copilot, your team can start from a foundation that's already been battle-tested by real investors and quants using it in production. They can focus on what makes your firm different: your data, your workflows, your intellectual property - instead of rebuilding what every other firm is also rebuilding.
121+
122+
But there's another advantage: OpenBB continues investing heavily in the foundational layer of Copilot (and workspace).
123+
124+
We're pouring capital into making it better, faster, and smarter - exactly as we have for over two years.
125+
126+
That ongoing investment benefits you directly. Your team gets access to improvements, new capabilities, and deeper financial domain understanding without having to fund the R&D themselves.
127+
128+
And because we're working across multiple firms and use cases, we're continuously learning what financial teams actually need to get done - which means the platform evolves with the industry, not in isolation. **For instance, we were the first workspace to support MCP server integration.**
129+
130+
This is how AI actually succeeds at investment firms: not by buying a generic tool, not by building from scratch, but by starting with proven infrastructure and layering their own innovation on top.
131+
132+
The firms that move fastest aren't the ones with the biggest budgets for AI, **they're the ones that don't waste their budgets on foundational problems that have already been solved.**
133+
134+
Source-available Copilot is that solve.
135+
136+
Your team gets over two years of R&D and a production-grade system. They focus on making it uniquely valuable for your investment process.
137+
138+
Everyone wins.
139+
140+
The question isn't whether your firm will use AI - it's whether you'll spend the next 24 months building what already exists, or the next 24 months actually innovating with it.
141+
142+
PS: Last week I shared an open source agent that I built over a weekend, our copilot is that + years of hard work. But if you want to get a glimpse to what you would get, I recommend starting from here: [Introducing OSOR Agent](https://didierlopes.com/blog/introducing-osor-agent)

0 commit comments

Comments
 (0)