Developers

Autonomy policy is code. Review it like a GxP change.

The Pildeon SDK is typed Python. Tools carry units and the approved design space; agents carry goals, autonomy levels and audit settings; the policy engine enforces all of it at call time. Every change is versioned and change-controlled.

Read the docs See the SDK

Typed Python SDKCLI and run APIDeterministic replay

run_b2c7e1 · agent graph · scenario 10 nodes · 1 QA approval gate
01 · orchestrator ingest.batch_record ✓ SUCCEEDED 02 · orchestrator twin.rehearse ✓ SUCCEEDED 03 · blend_dispense blend.uniformity ✓ SUCCEEDED 04 · granulate_dry granulate.dry ✓ SUCCEEDED 05 · compress_fill compress.press ✓ SUCCEEDED 06 · coat_dry coat.spray ✓ SUCCEEDED 07 · defect_inspect inspect.tablets ✓ SUCCEEDED 08 · quality_potency spectra.release_evidence ✓ SUCCEEDED 09 · orchestrator approve.qa ◆ APPROVAL 10 · quality_conf release.genealogy ✓ SUCCEEDED

Scenario: Run one immediate-release tablet batch through granulation, compression, coating and inspection inside the approved design space, with QA approval before release. A design walkthrough, not a customer record.

Developers

Define an agent, bound it, run it

The Pildeon SDK is typed Python. Tools are declared with schemas, units and design-space limits; the policy engine enforces them at call time, and every change to a tool or a limit is a change-control event.

What you get

  • Typed tool definitions with unit-aware ranges bound to the approved design space
  • Deterministic replay of any historical batch against a new model version
  • Local twin harness so a setpoint vector is rehearsed before it is shipped
  • Autonomy policy as code, versioned, reviewed and change-controlled like any other GxP change

Read the docs Developer guide

tp2_compression_agent.py 
# Bound the compression agent to tablet press 2's approved design space.
from pildeon import Agent, Tool, DesignSpace, Autonomy

force = Tool(
    name="press.write_setpoint",
    tags=["TP2.MAIN_FORCE", "TP2.FEED_FRAME_SPEED"],
    limits=DesignSpace.approved("family-A", "tooling-D"),
)

compress = Agent(
    id="agent.compress_fill",
    goal="weight, hardness on target; capping risk under threshold",
    tools=[force, Tool("press.read_force_curve", read_only=True)],
    # supervised: each move executes on approval
    autonomy=Autonomy.L2,
    # rehearse on the twin before every proposal
    verify="twin",
    # every call lands in the Part 11 audit trail
    audit="part11",
)

run = compress.start(line="L2", batch="B-240918")
for step in run.stream():
    print(step.name, step.status, step.duration)
CLI

Start a run from anywhere

The same run engine, the same policy checks, the same audit trail, from the terminal, the review console or the SDK.

pildeon · cli · scenario 
$ pildeon run "batch B-240918 on L2" --autonomy L2

→ plan composed          10 steps · 1 QA approval gate
→ twin.simulate          36 candidates · best #14 · capping risk under threshold
→ policy.evaluate        5 moves inside design space · release held for QA
→ executing              blend.uniformity .... ok   6m10s
→ executing              granulate.dry ...... ok   3m48s
→ executing              compress.press ..... ok   2m02s
→ executing              coat.spray ......... ok   4m25s
! approval required      batch_release · QA reviewer
→ approved               named QA reviewer · identity + timestamp logged
→ run complete           genealogy written · release package cited

$ pildeon runs show run_b2c7e1 --format genealogy
Capabilities

What the SDK gives you

Enough to define, bound, rehearse and audit an agent; not enough to bypass the policy engine.

Tools

Typed, unit-aware, bounded

Declare what a tool touches, its units, its limits from the approved design space and the autonomy level it needs.

Agents

Goal, tools, level, audit

Compose an agent from a goal, a tool set, an autonomy level, a twin verification step and an audit setting.

Twin harness

Rehearse locally

Run a candidate setpoint vector against the fitted twin before it is shipped; log the prediction with the run.

Replay

Deterministic

Replay any historical batch against a new model version and diff the recommendations, as evidence for change control.

Policy as code

Versioned, reviewed

Autonomy policy, allow-lists and limits live in version control and promote through the same change-control gate as models.

Audit API

Read the trail

Query the Part 11 audit trail by batch, agent, tool, reviewer or time; export in the formats QA and auditors ask for.

API

The run API

REST and server-sent events. Every endpoint is tenant-scoped and every mutation is audited.

Run API endpoints
EndpointPurposeReturns
POST /runsStart a run from a goal and a batch recordRun id, plan, approval gates
GET /runs/{id}Read a run recordSteps, statuses, durations, agents, tools
GET /runs/{id}/streamStream tool calls as they happenServer-sent events, in order
GET /runs/{id}/genealogyRead the batch genealogyTime-aligned states, actions, approvals, QC results
POST /approvals/{id}Record a reviewer decisionIdentity, decision, timestamp, signature
GET /modelsList models and versionsModel cards, validation status, pinned sites
POST /twin/simulateRehearse candidates on a fitted twinRanked candidates with predicted attributes
Run timeline · scenario

What the API returns

A run record is exactly this: ordered steps, each with agent, tool, duration, status and detail, plus the approval gate and the genealogy pointer.

run_b2c7e1 Line 2 · rotary tablet press · scenario SUCCEEDED
  1. 01 ingest.batch_record SUCCEEDED 1.2 s

    Plant Orchestrator · Pulled the master batch record, the approved design-space limits for this formulation and tooling family, and the site autonomy policy from MES; locked the target envelope for the run.

  2. 02 twin.rehearse SUCCEEDED 41 s

    Plant Orchestrator · Twindra simulated 36 candidate setpoint vectors on the as-made batch twin (granulation DEM, compression and ejection risk, coating CFD, dissolution surrogate) and ranked them on predicted content uniformity, capping risk and takt.

  3. 03 blend.uniformity SUCCEEDED 6 m 10 s

    Blend-and-Dispense · Spectryn scored in-line NIR spectra against the formulation's chemometric model; the blend reached its uniformity end point with confidence bounds inside the site acceptance limit and blending stopped on evidence rather than on a fixed time.

  4. 04 granulate.dry SUCCEEDED 3 m 48 s

    Granulate-and-Dry · Inlet air temperature and drying airflow trimmed inside the approved window to bring loss-on-drying to the target moisture band for compression; the move was bounded, logged and inside the design space.

  5. 05 compress.press SUCCEEDED 2 m 02 s

    Compress-and-Fill · Main compression force and feed-frame speed adjusted inside the approved window to hold tablet weight and hardness on target while the twin's capping-risk estimate stayed under the site threshold.

  6. 06 coat.spray SUCCEEDED 4 m 25 s

    Coat-and-Dry · Spray rate and inlet temperature re-phased against bed temperature and exhaust humidity to hold the film-coat window and avoid twinning and mottling; no excursion from the validated recipe envelope.

  7. 07 inspect.tablets SUCCEEDED continuous

    Defect-and-Inspect · Visqa classified chips, cracks, capping, spots, mottling and print defects on six synchronised cameras, within the sub-100 ms design budget, and linked a rising chip rate to coating spray rather than to tooling.

  8. 08 spectra.release_evidence SUCCEEDED 38 s

    Quality-and-Potency · Spectryn assembled the method-linked evidence chain for content uniformity and moisture with confidence bounds and citations to the validated analytical method, ready for QA review.

  9. 09 approve.qa APPROVAL 3 m 14 s

    Plant Orchestrator · Batch release is above every autonomy level Pildeon offers. The package was held for the named QA reviewer, who approved it in the review console; the decision, the reviewer identity and the timestamp are in the Part 11 audit trail.

  10. 10 release.genealogy SUCCEEDED 52 s

    Quality-and-Conformance · Conforma wrote the time-aligned batch genealogy (every sensor state, recommendation, approved action, equipment response and QC result) and the cited release package to the immutable audit trail.

Tool calls · scenario

Streaming a run

Every tool invocation, argument and result is written to an immutable, human-readable log, and every reasoning step is expandable. Nothing about a run is hidden from the plant or from QA.

tool-call stream · run_b2c7e1
  1. 06:12:04mes.read_batch_record({ line: "L2", product: "IR tablet · family A" })
  2. 06:12:05ok design space loaded · autonomy policy L2 · QA reviewer assigned
  3. 06:12:06twin.simulate({ candidates: 36, horizon: "full batch" })
  4. 06:12:47ok candidate #14 selected · predicted CU inside limit · capping risk under threshold
  5. 06:12:48policy.evaluate({ candidate: 14, autonomy: "L2" })
  6. 06:12:48ok 5 setpoint moves inside design space · release held for QA
  7. 06:12:49nir.score({ probe: "blender-1", model: "cu-famA-v3" })
  8. 06:18:59ok blend end point reached · confidence bounds inside acceptance limit
  9. 06:19:00plc.write_setpoint({ tag: "FBD1.INLET_T", within: "design space" })
  10. 06:22:48ok loss-on-drying in target band · move logged to audit trail
  11. 06:22:49press.write_setpoint({ tag: "TP2.MAIN_FORCE", within: "design space" })
  12. 06:24:51ok weight and hardness on target · capping risk under threshold
  13. 06:24:52vision.stream({ cameras: 6, model: "visqa-defect-v2" })
  14. 06:27:10chip rate rising on camera 4 · attributed to spray rate, not tooling
  15. 06:27:11coater.write_setpoint({ tag: "CT1.SPRAY", within: "design space" })
  16. 06:29:14ok chip rate back to baseline · film window held
  17. 06:29:15nir.evidence_chain({ attributes: ["cu", "moisture"], method: "validated" })
  18. 06:29:53ok evidence chain assembled · citations attached
  19. 06:29:54approval.request({ action: "batch_release", reviewer: "QA" })
  20. 06:33:08ok approved by the named QA reviewer · identity and timestamp logged
  21. 06:33:09audit.write_release_package({ batch: "B-240918", genealogy: "full" })
  22. 06:34:01ok scenario complete · genealogy written · package cited

Reasoning trace

Plan, thought, action, observation, exactly as the orchestrator records it.

GoalRun one immediate-release tablet batch through granulation, co…

Run one immediate-release tablet batch through granulation, compression, coating and inspection inside the approved design space, with QA approval before release.

PlanRehearse on the twin first. Stop blending on NIR evidence rath…

Rehearse on the twin first. Stop blending on NIR evidence rather than on a fixed time. Trim drying, compression and coating only inside the approved design space. Inspect continuously. Hold release for the named QA reviewer, always.

ThoughtThe fixed recipe blends for a set time regardless of what the …

The fixed recipe blends for a set time regardless of what the spectra say. On the twin, the candidate that stops on evidence and trims inlet temperature before compression keeps content uniformity and moisture inside their limits with the lowest predicted capping risk.

Actiontwin.simulate(candidates=36) → candidate #14: evidence-based b…

twin.simulate(candidates=36) → candidate #14: evidence-based blend end point, bounded drying trim, compression force held inside the window, spray rate re-phased to bed temperature.

ObservationEvery candidate that pushed turret speed for takt raised predi…

Every candidate that pushed turret speed for takt raised predicted capping risk above the site threshold and was discarded. Rate is never traded for a quality attribute.

ActionExecute candidate #14 under autonomy level L2: five setpoint m…

Execute candidate #14 under autonomy level L2: five setpoint moves inside the design space, each logged. Batch release routed to QA.

ObservationChip rate rose on camera 4 at 06:27:10. Visqa attributed it to…

Chip rate rose on camera 4 at 06:27:10. Visqa attributed it to coating spray rather than tooling; a bounded spray-rate trim returned it to baseline in about two minutes.

OutcomeScenario complete. Every move inside the approved design space…

Scenario complete. Every move inside the approved design space, the QA approval recorded with identity and timestamp, and the cited release package written by Conforma to the immutable audit trail.

Workflow

From idea to a line, safely

Development in a regulated plant has a shape, and the SDK follows it.

  1. Define the tool and the agent in code, bound to the approved design space.
  2. Rehearse on the twin harness; log predicted attributes.
  3. Replay against historical batches; diff against the current model.
  4. Review the change with MSAT and QA; golden-dataset evaluation gates the merge.
  5. Promote under change control to shadow mode on the line; nothing writes.
  6. Graduate the autonomy level only when the site decides, and record who decided.
Autonomy

Four levels, set per agent and per setpoint

A tablet plant does not go from fixed recipes to autonomy in one step, and validated parameters never change without sign-off. Pildeon makes the level explicit, auditable and reversible at any time.

Autonomy levels and the human role at each
LevelWhat the agent doesWhat the human doesTypical time to reach
L0 · ShadowObserves, scores and predicts; writes nothingRuns the line as today; compares predictions to QC resultsWeeks 1 to 6 of a design partnership
L1 · AssistRecommends setpoint moves and explains whyProcess engineer enters the move; QA reviews the logAfter shadow-mode accuracy is shown
L2 · SupervisedProposes a bounded move; it executes on approvalApproves each move in the review consoleAfter assist-mode ROI is shown
L3 · BoundedMoves low-risk compression, coating and inspection setpoints inside the approved design spaceHandles exceptions; approves every releaseAfter validation and change-control review

There is no unattended level. Batch release and any move outside the approved design space always wait for a named QA reviewer, and every model or prompt change runs under GxP change control.

Multi-agent handoff

Agents negotiate, they do not collide

Two agents will want the same setpoint. The orchestrator arbitrates on the run goal and the quality attributes, not on who asked first, and the handoff is logged like any other step.

Contested move: turret speed on tablet press 2

  1. 01yield_takt.request(turret +4%) SUCCEEDED0.2 s

    Yield-and-Takt wants more turret speed to recover takt lost to an upstream changeover.

  2. 02quality_potency.request(hold) SUCCEEDED0.2 s

    Quality-and-Potency wants turret speed held: the twin's content-uniformity and capping-risk estimates rise with dwell time lost.

  3. 03orchestrator.arbitrate SUCCEEDED0.4 s

    A quality attribute outranks takt under every Pildeon run goal. Quality-and-Potency wins the setpoint; the takt request is declined and the reason recorded.

  4. 04yield_takt.replan(campaign) SUCCEEDED6 s

    Yield-and-Takt recovers the schedule by re-sequencing the next changeover with cuOpt instead of pushing the press.

Arbitration rules

  • Quality attributes first. Content uniformity, potency, dissolution and defect rate outrank yield, energy and takt in every arbitration.
  • The design space is a wall, not a target. No agent can request a move outside the limits the site's validation approved.
  • Time-boxed ownership. An agent holds a contested setpoint for a bounded window, then must re-justify.
  • Everything is logged. The losing request, the score and the reason all appear in the batch genealogy.
  • Deadlocks escalate to a human rather than resolving by timeout.

See the agent roster

Integrations

It speaks tablet plant, not cloud

Pildeon reads and writes through the systems already on the floor: presses, coaters, granulators, PAT probes, MES, LIMS and historians. No rip-and-replace, no parallel historian. Connectors are built with design partners against their stacks; the vendors listed are the ones common in OSD plants.

Blending and granulation

GEA, Glatt, Syntegon, Freund-Vector, L.B. Bohle

Blender, granulator and fluid-bed telemetry; bounded setpoint writes

Tablet presses and fillers

Fette Compacting, Korsch, IMA, Syntegon, MG2

Force curves, fill depth, ejection force; bounded setpoint writes

Coating systems

GEA, Glatt, O'Hara, Thomas, Freund

Spray, inlet, bed temperature, exhaust humidity; bounded writes

PAT and NIR

Bruker, Sartorius, Viavi probes and analysers

Spectra across blend, granulation, drying and compression

Vision

Line cameras and existing inspection systems

Frames, defect maps, reviewed images

Pharma MES and eBR

Körber Werum PAS-X, Rockwell, Siemens Opcenter

Master batch records, campaigns, batch genealogy

LIMS and historians

Site LIMS, OSIsoft PI and equivalent historians

QC results, time-series backfill and replay

Identity and edge

Azure AD, Okta, LDAP · NVIDIA Jetson and on-prem GPU

SSO, RBAC, named approvers · inference at the line

See all integrations

Guardrails

An agent that can move a compression force needs a leash

Pildeon writes to production equipment in a GMP plant. Every capability is scoped to the approved design space, every write is policy-checked, every action is written to an immutable Part 11 audit trail the site owns, and no agent can release a batch.

  • Bounded action space. Each agent can only write to an explicit setpoint allow-list, inside the limits the site's validation approved.
  • Policy engine before every write. Autonomy level, design-space limits, batch state, model version and reviewer assignment are all evaluated before a setpoint moves.
  • Human-in-the-loop gates. Batch release, any move outside the design space and any change to a validated parameter wait for a named QA reviewer.
  • Immutable audit trail. Append-only, hash-chained, exportable, retained on the site's own storage, with reviewer identity and timestamp on every decision.
  • Hard fallback. Loss of the edge node, the network or the model returns the line to the equipment's last known-good recipe.
  • Tenant and IP isolation. Formulations, recipes and models never cross a customer boundary. Validated on-prem deployment is available for regulated producers.

Compliance posture

StandardScopeStatus
21 CFR Part 11Audit trail, e-signatures, records RUNNING Designed in from day one
GAMP 5Validation package for the platform RUNNING In progress, planned for the first design partners
cGMP · ICH Q8 to Q13Design space, QbD, continuous manufacturing RUNNING Design-aligned; site validation decides
SOC 2 Type ICloud control plane QUEUED Planned, first six months
SOC 2 Type IICloud control plane QUEUED Planned, months six to twelve
Data residencyValidated on-prem, air-gapped option SUCCEEDED Available by design

Read the security and validation overview

FAQ

Straight answers

The questions site directors, MSAT engineers and QA leads actually ask in the first meeting.

Yes, but only inside an explicit setpoint allow-list bounded by the design space your validation approved, and only at the autonomy level your site has set. Level 0 is shadow mode: Pildeon observes and predicts and writes nothing. Most sites spend their first weeks there.

Get started

Build on the run engine

SDK access comes with the design partnership. Tell us the line and the workflow and we will set up the harness.

Book a line review Documentation