Studio/2.0
2.1Agents
Configure existing agents
Modify pre-built agents through modular system prompt parts that work both interactively and autonomously. Connect your data sources, give them access to the tools they need, and configure skills for the way your teams work.
Generate new agents
Create new agents for your specific purpose with no code. Each generated agent is set up the same way: modular, configurable, and ready for you to refine for your needs.
Test, observe, and optimise
Test your agents and observe their end-to-end behaviour. Evaluate the results, then refine and optimise for efficient, cost-effective, and sustainable models suited for each task.
2.2Applications
Build or access pre-built
Build real enterprise AI applications from scratch or access pre-built applications ready to deploy across your organisation and teams.
Real-time data and decisions
Give your applications access to real-time data. Turn operational data into real-time decision making and autonomous execution at scale.
Deploy across your organisation
Deploy applications to the teams that need them. Each application is powered by a configured agent with the UI, tools, and access it needs.
2.3Resources
Your data and knowledge
Bring documents, standards, and domain knowledge into shared resources. Every agent and workflow starts from the context your business already has.
Your tools and connections
Connect the systems and services your teams rely on. Each integration has explicit permissions and access boundaries built in.
Your way of working
Encode styles, skills, and reusable guidance once. Agents inherit them across every task so results reflect your standards, not generic defaults.
2.4Models
Your models, your choice
Bring hosted, sovereign, or self-run models. Configure them per agent and per task, and evaluate against the work your teams actually do.
Switch without rebuilding
Models are a configurable part of the platform, not its foundation. Change providers or endpoints and the business context stays.
Optimise for cost and performance
Balance quality, latency, cost, and data handling per workflow. Keep those decisions visible and revisit them as models evolve.
2.5Observability
End-to-end run records
Follow every step of a run: tool calls, agent decisions, intermediate outputs, and the final result. Nothing hidden.
Cost and performance
Token usage, duration, and cost per run. Find the expensive or slow steps and decide what to optimise.
Errors and diagnostics
See which runs succeeded, failed, or completed with warnings. Trace each issue back to the exact step that caused it.
2.6Evaluation
Evaluated automatically
Every run is scored as it completes: success, cost, duration, and turns, broken down by model, agent, and workflow. No manual benchmarking.
Recommendations, not guesswork
The system spots what can be improved: a leaner instruction, a smaller model that does the job, or a tool that replaces a step the agent keeps repeating. It recommends the change with the evidence behind it.
Modular down to the smallest part
Instructions, skills, styles, tools, and models are separate parts, so each can be changed on its own. Every change is measured version against version on your own work. Keep what performs, roll back what does not.