OMNI Agent
More than a chatbot. Closer to a real operator.
OMNI Agent is the work layer in the stack. It combines a purpose-built prompt system, routed skills, structured tools, memory, and product context so it can do investor work instead of just narrating investor work.
Why it exists
A generic model can answer. An investor agent has to continue the work.
The hard part is not generating a sentence. It is keeping the answer tied to source-backed inputs, choosing the right workflow, preserving context from the day, and returning something that survives handoff. That is the job OMNI Agent is built for.
- It starts from OMNI Datastream when the work should be grounded in filings, ownership, statements, or market utilities.
- It routes into built-in skills when the task has a better playbook than a blank prompt.
- It carries product context when it is embedded inside OMNI Pulse.
How the harness works
A cleaner path from prompt to useful result.
This layer reduces improvised reasoning when the system already has a better path available.
Step 1
Start from grounded inputs
Datastream, market data, documents, spreadsheets, and product context give the agent cleaner material than a raw web scrape.
Step 2
Route into the right skill or tool
The harness can pick from built-in skills and structured tools so repeated tasks stop behaving like first-time tasks.
Step 3
Keep the source trail
Outputs are expected to preserve request IDs, provenance, freshness, and supporting artifacts when the workflow needs them.
Step 4
Return something you can keep using
The point is not a clever paragraph. The point is a brief, export, table, or next action that fits the day.
What the harness includes
The stack behind the agent.
Prompt stack
OMNI Agent uses a prompt system designed around investor work, not generic chat pleasantries.
Skill routing
The agent does not have to guess the whole job from scratch every turn. It can route into built-in skills and workflows.
Structured tools
Datastream requests, market data, documents, spreadsheets, and agent-side workflow tools all live in the same loop.
Context-aware chat
Inside Pulse, the agent can reason from the page, symbol, portfolio, or feed the user is already looking at.
Artifact-friendly outputs
The agent can produce briefs, tables, exports, and structured outputs that survive handoff instead of vanishing into chat history.
Portfolio memory
Repeated patterns, preferred views, and ongoing research can stay in working memory instead of being restated every time.
Evaluation posture
OMNI measures skill and workflow quality against real tasks instead of demo polish.
Autoresearch loops
The research system can refine, score, and improve repeated workflows instead of staying static.
Multi-surface delivery
The same intelligence layer can show up in chat, CLI, SDK, API, and embedded product surfaces.
Cross-device preferences
Thinking level, input behavior, and agent personalization sync across web, desktop, and CLI without manual re-entry.
Who the skills are for
The skill inventory is broad on purpose.
Product-facing skills
35 skills help with research, filings, modeling, news, monitoring, and the daily market loop inside OMNI Pulse.
Developer-facing skills
2 skills support prompt scaffolding, repo-native workflows, and code or spreadsheet tasks that need a more exact operating mode.
Operator-facing skills
1 skills connect the agent to support, inbox, and system workflows that matter after the first answer ships.
Three investor playbooks
The same agent, different jobs.
Macro, global, and U.S. equity workflows lean on different parts of the stack. OMNI Agent is built to switch modes without pretending they are all the same task.
Macro investor
Use OMNI Agent to move quickly between rates, curves, inflation prints, volatility, and market reactions without losing the source chain.
Macro and treasury data skills
Pull FRED, rates, treasury, and curve context into the same working thread.
Volatility and regime context
Use OMNI Volatility Score and live market surfaces to frame regime change faster.
Briefings and inbox follow-through
Turn a macro shift into a daily briefing note, an alert review, or a portfolio question.
Built-in skills
Thirty-eight active skills. Public-safe. Actually useful.
This is the current built-in OMNI Agent skill inventory from omni-apps. The point is breadth with structure, not a random bag of prompts.
Research skills
Skills that help OMNI Agent build context around one company, one theme, or one peer set quickly.
product
Company Primer
Builds the investor-grade 'what does this company do and why does it matter' memo.
company-primer
product
Competitors
Finds direct peers, substitutes, and important adjacent players with a reason for each inclusion.
competitors
product
Earnings Preview
Builds a pre-earnings brief around expectations, key KPIs, and what can actually move the story.
earnings-preview
product
Earnings Recap
Turns a fresh quarter into a scorecard, a reaction read, and a faster first-pass summary.
earnings-recap
product
Estimate Analysis
Pulls consensus expectations, ratings, and target posture so the agent can explain what the market is leaning toward.
estimate-analysis
product
Industry Primer
Builds the market-structure and profit-pool context around a sector before the company-level work begins.
industry-primer
product
Investment Thesis
Builds a clear bull case, bear case, catalyst map, and risk frame from the same grounded inputs.
investment-thesis
product
Deep Research (last30days)
Searches the last month of web, social, and media chatter when fresh outside-the-filing context matters.
last30days
product
Multi-Company Analysis
Normalizes a universe and compares several companies without drifting into inconsistent one-offs.
multi-company-analysis
product
Read News
Filters the daily news feed down to what matters for a ticker, theme, or watchlist.
read-news
product
Upcoming Events
Surfaces the next earnings, dividends, splits, IPOs, and macro events before they surprise the workflow.
upcoming-events
Trust and grounding
The agent is strongest when it can show what it used.
OMNI Agent is designed to keep the source trail intact when the work depends on Datastream, market data, files, or product context. The point is not to sound convincing. The point is to stay legible after handoff.
Where it shows up
One intelligence layer, several entry points.
OMNI Pulse chat
Ask about the page, the portfolio, the briefing, or the market context you are already looking at.
API and SDK surfaces
Call the same harness from products, services, automations, or evaluation runs.
CLI workflows
Use `omni chat` from the terminal when the fastest path is still a keyboard. Auth, streaming, permissions, and session history carry over.
Embedded workspace tools
The agent can work alongside symbol pages, terminal routes, and briefings instead of outside them.
Artifacts and exports
Return research that can be saved, passed around, and used again.
Memory that compounds
Repeated workflows get better when the system remembers what mattered last time.
Use it
Talk to it in the product. Or wire it into your own stack.
OMNI Agent is designed to be useful both inside the workspace and through programmatic surfaces.