Code-first agent workflow runtime console

For

Engineering teams building and running AI agent workflows that use tools to complete multi-step tasks

Solves

Agent workflows are spread across separate tools for definition, execution, memory, scheduling, permissions, grading and observability, so teams rent several subscriptions and still cannot see or control a run end to end.

Delivers

A source-linked assistant and administrator console

Built in

about 4 weeks of creation time, MVP in 4 days

Investment

$14,000 for the MVP, $47,500 for the full product

Run it

Inside your business, or as part of your offer to clients

Reduce the number of rented tools and give the team one owned runtime for defining, running, inspecting and approving agent workflows.

1. Define agents and workflows in code or a text-based DSL inside the codebase.
2. Plan steps and execute tools with retries and conditional flows.
3. Run workflows on a durable runtime with retries, cancellation and replay.
4. Persist short-term and long-term memory with callbacks across sessions.
5. Carry state across clean contexts using files.
6. Connect agents to external tools and APIs.
7. Schedule agent runs.
8. Run agents from the command line in different contexts.
9. Restrict agent access to explicitly allowed tools and scopes.
10. Start a fresh worker and critic session each cycle so no prior conversation carries over.
11. Grade against a markdown rubric and produce fix notes.
12. Show cycle progress, logs and state in a local web dashboard and allow cancelling a run mid-cycle.
13. Bridge ChatGPT to send goals to the daemon and monitor runs without exposing the daemon publicly.
14. Switch the critic to an open-ended improve-or-ship question after a PASS.
15. Disable network access inside the project folder during runs.
16. Pause workflows for human approval at defined points.
17. Parse a workflow to list tools, guards and approval pauses without executing it.
18. Wire in tracing and cost reporting through Langfuse.
19. Enforce separation so the DSL cannot import or call into the host application.
20. Store workflows as plain text files for Git history, PRs and readable diffs.

Everything these tools do, in one app

• Code-first workflow definitions Define agents and workflows directly in code or a text-based DSL within your codebase.Found in OpenMolt, Finyuus
• Planning and execution loop Agents plan steps and execute tools to complete multi-step tasks with retries and conditional flows.Found in OpenMolt
• Durable execution Workflows run with retries, cancellation, and replayability backed by a durable runtime.Found in Finyuus
• Memory persistence Short-term and long-term memory with persistence callbacks carry state across sessions.Found in OpenMolt
• File-based memory State is carried across clean contexts using files.Found in AgentLoop
• Tool and API integrations Connect agents to external tools and APIs.Found in OpenMolt
• Scheduling Schedule agent runs.Found in OpenMolt
• CLI runner Run agents from the command line in different contexts.Found in OpenMolt
• Capability-based permissions Restrict agent access to explicitly allowed tools and scopes.Found in OpenMolt
• Fresh worker and critic per cycle Each cycle starts new Codex sessions for worker and critic so no prior conversation carries over.Found in AgentLoop
• Rubric-based grading Define pass/fail criteria in a markdown file that the critic grades against and produces fix notes.Found in AgentLoop
• Live dashboard A local web interface shows cycle progress, logs, and state, and allows cancelling a run mid-cycle.Found in AgentLoop, Finyuus
• MCP bridge Allows ChatGPT to send goals to the daemon and monitor runs without exposing the daemon publicly.Found in AgentLoop
• Polish mode After a PASS, leftover cycles switch the critic to an open-ended improve-or-ship question for refinements beyond the rubric.Found in AgentLoop
• Network disabled in project folder Runs are contained by disabling network access inside the project folder.Found in AgentLoop
• Human approvals Workflows can pause for human approval at defined points.Found in Finyuus
• Static analyzability Parse a workflow to see which tools it calls, which guards it runs, and where it pauses for human approval, without executing it.Found in Finyuus
• Integrated observability Tracing and cost reporting are wired in via Langfuse.Found in Finyuus
• Separation of AI logic from application code The DSL cannot import or call into the host application, enforcing separation.Found in Finyuus
• Git-diffable text workflows Workflows are stored as plain text files, giving Git history, PRs, and readable diffs.Found in Finyuus

What goes in, what comes out

The workflow

1. In
   Start with

   Workflow source files, tool and API credentials, memory stores, rubrics and permission scopes

2. 1

   Confirm the buyer's problem and scope

3. 2

   Collect workflow source files

4. 3

   Tool and API credentials

5. 4

   Memory stores

6. 5

   Rubrics and permission scopes

7. 6

   Then follow this sequence: 1

8. Out
   Finish with

   A source-linked assistant and administrator console

AI does the heavy lifting, people stay in charge

Use AI to interpret permitted inputs, suggest structured mappings and generate candidate outputs for the three stated task modules. Use deterministic code for arithmetic, schema validation, hard constraints and reproducible tests. Review source-linked explanations and uncertainty before accepting results. One fixed runtime version and approved tool set; final code review and release decisions remain engineering. A model suggestion is never a verified fact, professional decision or authorization to act.

What your team sees

Primary screens: Workflow source and definitions, Live run dashboard, Review and approval queue. Use a file tree for workflow text files, a central editor with diff view, and a right-hand panel for tools, guards, memory and approval points. Let users compare run versions side by side. Display queued, running, paused, passed and failed states. Provide a client preview link with comments anchored to the relevant run step. Make the task-specific outcome a source-linked assistant and administrator console visible beside its evidence, review state and value baseline.

Accounts and administration

Project ownership, workflow versions, run history, approval states, tool scopes, schedule limits, download history and a rights record for supplied material. Add organization access boundaries, named reviewers, usage caps, data retention controls, export logs and explicit approval for external actions.

Integrations and data access

Team-owned repositories, authorized tool and API endpoints and permitted model providers. Cloud run storage, Git hosting and observability destinations. Start with file exchange and validate destination specifications before promising direct deployment. Start with authorized file exchange. Validate current provider access, usage rights and schema behavior before promising a connector.

We build with our own AI software development factory, so most implementations take days to a few weeks of creation time, not months. You see working software at every step, and exact timing depends on availability.

1. 1

   Scoping call

   Day 1

   Thirty minutes on your process, your data and how you want to run it: for your own team, or for your clients. You get a fixed scope and price for the MVP.

2. 2

   MVP

   4 days

   One buyer segment, one recurring use case; first modules: define agents and workflows in code or a text-based DSL inside the codebase; plan steps and execute tools with retries and conditional flows. Manual review in the loop. Built by our AI software factory.

3. 3

   Paid pilot

   5 days

   Accounts, roles, review states, audit trail and the first integration, hardened for two to three paying pilot customers.

4. 4

   Full product

   9 days

   Self-serve onboarding, billing, monitoring and the wider integration set.

5. 5

   Run and improve

   Monthly

   We host, monitor and improve it for a fixed monthly fee, or hand it over to your team. How the retainer works.

Why we start with an MVP

An MVP, or minimum viable product, is the smallest version that your users can actually work with. It is not a cheap version of the full solution. It is a test, built to answer the questions that decide whether the rest is worth building.

1. Pick the riskiest assumption. Here: will engineering teams building and running AI agent workflows that use tools to complete multi-step tasks use it to solve "agent workflows are spread across separate tools for definition, execution, memory, scheduling, permissions, grading and observability, so teams rent several subscriptions and still cannot see or control a run end to end"?
2. Build only what tests it. One team, one use case, a few core modules. People do the rest by hand for now.
3. Run a paid pilot. Agree quality and outcome thresholds before the pilot using this measure: Completed multi-step tasks per engineering hour and rework after a run is accepted.
4. Measure, then decide. Track completed multi-step tasks per engineering hour and rework after a run is accepted; accepted-output rate; material error rate; reviewer correction time; actual repeat purchase. Then expand, change course or stop, with evidence instead of opinions.

MVP scope for this solution. Pilot scope: One fixed runtime version and approved tool set; final code review and release decisions remain engineering. Implement one approved input format, a bounded representative case set and the first two task modules: define agents and workflows in code or a text-based DSL inside the codebase; plan steps and execute tools with retries and conditional flows. Support the third module with operator review: run workflows on a durable runtime with retries, cancellation and replay. Include source references, corrections, basic organization access, approval states, export and value measurement. Use managed operator assistance for unresolved exceptions. The cost estimate covers this narrow prototype, not unrestricted multi-tenant scale, complex production integrations, specialist certification or physical operations.

After the MVP. Once paid pilots prove usefulness, automate repeatable reviewed steps and add one verified source integration. Expand supported inputs and case volume only after new evaluation cases pass. Build reusable customer configurations and recurring value reports around a source-linked assistant and administrator console. Retain the explicit scope boundary: One fixed runtime version and approved tool set; final code review and release decisions remain engineering.

What the build depends on. Workflow upload and preview, asynchronous run jobs, editable version history, reviewer access and tested export formats. High-fidelity production requires specialist engineering QA. Obtain representative authorized cases, baseline measurements, qualified reviewers and a buyer-side decision owner. Specific limitation: One fixed runtime version and approved tool set; final code review and release decisions remain engineering.

A planning range to start the conversation, not a quote. You pay per phase, so you can stop after the MVP.

Running costs per month

A rough indication of monthly hosting and AI model costs once it is live, not tested. Real costs depend on usage, file sizes and the models chosen.

Your brand, or this one

Run it under your own brand, or start from this concept style.

Selling it to your own clients: the go-to-market playbook

Pricing to test

Test a USD 300-1,500 fixed pilot for one defined workflow package. Offer a monthly production allowance after repeat demand. Quote complex multi-tenant or specialist integrations separately. These are test prices, not market benchmarks. Package the initial sale as one bounded source-linked assistant and administrator console. Recurring fees must specify volume, review depth and integration support. For exchanges, test a disclosed coordination or successful-service fee rather than holding customer funds. Reprice only after measuring real delivery labor; platform-build cost is separate from a commercial pilot fee.

Message to test

Reduce the number of rented tools and give the team one owned runtime for defining, running, inspecting and approving agent workflows. Demonstrate a concrete source-linked assistant and administrator console using the buyer's approved example and show the baseline, corrections and actual delivery effort.

Where to find buyers

Engineering teams building and running AI agent workflows professional communities; specialist consultants serving this buyer; permissioned partner introductions; practical demonstrations at relevant trade or practitioner events.

Lead magnet

A reviewed sample source-linked assistant and administrator console from a small authorized input set, with a transparent calculation of completed multi-step tasks per engineering hour and rework after a run is accepted and no promised savings.

The first 30 days

1. Week 1: interview five engineering teams building and running AI agent workflows and inspect a recent example of agent workflows spread across separate tools for definition, execution, memory, scheduling, permissions, grading and observability.
2. Week 2: prepare a consented or synthetic demonstration of the three task modules.
3. Week 3: seek one bounded paid pilot with agreed baseline and acceptance criteria.
4. Week 4: measure completed multi-step tasks per engineering hour and rework after a run is accepted, reviewer effort and repeat-purchase interest. This is a demand-validation plan, not a thirty-day full-product delivery promise.

Paid pilot

Agree quality and outcome thresholds before the pilot using this measure: Completed multi-step tasks per engineering hour and rework after a run is accepted. Continue only if the buyer accepts the actual output, the intended job outcome improves without unacceptable errors, and measured delivery cost fits willingness to pay. Revise or stop if access is unavailable, qualified review cannot be provided, or apparent savings disappear after corrections and support. Use held-out cases when comparing model quality; use a properly reviewed comparison design before making causal claims. Record missing cases and negative results alongside successful outputs.

Success metrics

Completed multi-step tasks per engineering hour and rework after a run is accepted; accepted-output rate; material error rate; reviewer correction time; actual repeat purchase.

Retention and expansion

Repeat the workflow when the buyer again needs a source-linked assistant and administrator console. Retain permissioned settings and reviewed examples, report realized value honestly, and sell increased volume or adjacent approved workflows only after contribution margin and quality remain acceptable.

Why clients would pick it

A reusable library of approved workflow patterns, tool scopes and review examples, together with reliable delivery for a narrow engineering niche. Build a permissioned library of representative task cases, reviewer corrections and verified operating constraints for engineering teams building and running AI agent workflows that use tools to complete multi-step tasks. Repeatable delivery and useful integrations matter more than access to a base model.

Alternatives and positioning

OpenMolt, AgentLoop and Finyuus, plus generic orchestration libraries and internal scripts. Compare this product with the buyer's present method on completed multi-step tasks per engineering hour and rework after a run is accepted. Offer a bounded paid workflow instead of claiming broad autonomous expertise. Market uniqueness and competitor coverage are not verified.

Main delivery costs

Model calls, tool and API usage, storage, reviewer hours, client revision rounds and licensed source assets. Additional initial validation requires representative authorized sample preparation, buyer interviews, buyer-side evaluation and bounded validation of a source-linked assistant and administrator console. Track cost per accepted output, including correction work, unsuccessful cases and support.

Preserve code ownership, source attribution, credential handling and usage permissions. Engineering owners approve substantive changes and deployment scope. One fixed runtime version and approved tool set; final code review and release decisions remain engineering. Keep all consequential actions under authorized human control and do not fabricate missing inputs, permissions, professional judgments or market evidence.