Manual Software
Effective agents need more than a model.
Manual Core checks an agent’s actions before they take effect. Index gives your coding agents a shared project to work from. Both grew out of our approach to AgentUX, designing for the agent as a user.
- What do I need
to know? - Does my work
hold up? - What should
I change?
Manual Core
Authorize an agent’s next action in context.
Manual Core uses semantic authorization for real-time evaluation and permission granting. Its guardrails check whether the evidence supports an agent’s next action. When it doesn’t, the agent gets feedback it can act on. Those safety checks help teams trust agents with work like sending a customer a quote on approved terms.
- Process graphs
- Connect an action to the evidence and dependencies the agent needs. Both the agent and its tools can refer to the same node in the process graph.
- Tool-side pre-authorization
- Before a tool acts, it can check the agent’s permissions, required evidence and calculations.
- Semantic response checks
- Check tool results and compare a proposed reply with trusted examples. Authorize delivery, return feedback for revision, or route the reply for review.
- Delegated identity and audit
- Follow the steps and tool calls together so you can trace how an agent reached a decision.

The discount was approved for annual prepayment. The draft offers monthly billing. Before sending, the quote tool asks Core whether the terms match.
Recovery after a denial
The tool holds the quote and tells the agent which term differs. The agent can revise the payment schedule or request approval for monthly billing, then submit the quote for the same check.
Index by Manual Software
Harness engineering for everyone.
Ask your agents to build a feature from the discussion in an Index project. Index is the foundation for that collaboration. They can pick up the decisions you made together, follow your team’s workflow and review each other’s work.
- Reusable workflows
- Agents can create a way of working as a project takes shape. A team can reuse it on the next job or switch workflows while work is underway.
- Coordination across sessions
- Agents can find ready tasks, claim one and ask a teammate to check the result.
- Lightweight memory
- Decisions, artifacts and checkpoints stay with the project. A fresh session in another tool can pick up where the team left off.
Free for companies with fewer than 50 people. 50+ people? Contact us

- Claude CodeThe tester records the stale results
- CursorThe fixer corrects the filter reset
- CodexThe reviewer retests both search cases