A contract lifecycle platform that reads every clause, and stands behind it.
Bind a SharePoint folder. CLMSpace extracts every obligation with verbatim citations anchored to the source PDF, runs each draft against a scoped playbook (User → AI → Market Practice), and ships counter-positions, renewal packs, and deviation flags with a human-in-the-loop verification gate.
Runs inside your Microsoft 365 tenant · per-tenant Dataverse · UK / EU / US data residency · SOC 2 Type 1 in flight
Liability cap — clause 11.2
Capped at 12 months’ fees; uncapped for confidentiality & IP indemnity.
p.14 · bbox 142,318 → 480,360
MSA · vendor · UK
Aggregate cap = 100% fees paid in prior 12 months; carve-outs for confidentiality, IP, mandatory law.
User-verified · Tier 1
Acme MSA · flagged
Uncapped indirects vs. playbook. Suggested redline drawn from your clause library.
High risk · copy redline · copy email
How it works
From SharePoint folder to scoped negotiation, in three steps.
No parallel document store. No Power Automate flows to babysit. Your contracts stay where they live; the platform reads them, verifies them, and runs every draft against your playbook.
- Step 1
Bind your folders
Point CLMSpace at the SharePoint libraries you already keep contracts in. The /Templates folder seeds your User-tier playbook; /Agreements is watched on a 5-minute Microsoft Graph delta cursor. .pdf, .docx, .md and .txt all supported. Ingest backfills existing files on bind.
- Step 2
Extract & verify
Claude Haiku does tool_use extraction into a strict obligation schema — ~$0.001 per contract. Every obligation arrives with a verbatim source quote, page number, and PDF bounding box you can click to open the source clause. AI-extracted rows hold as 'unverified' until an admin signs off; the gate blocks downstream connectors until they do.
- Step 3
Playbook drives the draft
Every clause is resolved against your scoped playbook — scoped by agreement_type × party_role × jurisdiction. Tier 1 (User), Tier 2 (AI from your templates & history), Tier 3 (Market Practice corpus). Deviations flag in the grid; the Negotiator drafts counter-positions from your clause library; the Drafter assembles renewal packs.
What your legal team is doing instead of negotiating.
Four jobs that disappear into spreadsheets, calendar invites, and Word documents that don’t talk to each other. Every one of them is downstream of an AI that should have read the clause for you.
Hundreds of PDFs in SharePoint. “What is the longest confidentiality tail in our active NDAs?” takes three days and a junior associate. A register that lists agreement names is not the same as a register that knows their clauses.
An MSA auto-renews with a CPI+3% uplift no-one negotiated out in 2023. Notice to terminate expired 14 days ago. Calendar reminders aren’t lifecycle management; you need the standard-vs-actual delta on the desk 60 days early, not the expiry on the desk 7 days late.
“Our cap is 12 months” — until seven counter-paper MSAs slipped through with uncapped indirects because the redline review was rushed. The standard exists; it’s just not enforced anywhere a machine can check.
Same indemnity argument re-litigated every quarter. Junior counsel relearning what senior counsel already settled. Your clause library is in Tom’s head — the negotiator agent needs it in a database.
The playbook authority model
One playbook. Three tiers of authority. Scoped to the deal.
Every counter-position, deviation flag, and renewal draft resolves against the same three-tier playbook — scoped by agreement_type × party_role × jurisdiction. User overrides AI. AI overrides Market Practice. Nothing overrides what your General Counsel signed off on this morning.
What you have agreed is house view.
- Manually entered by counsel
- Imported from your CSV / register
- AI-derived rows you verified into Tier 1
Authority: Highest. Quoted verbatim in every draft.
What your templates and history say you do.
- Extracted from a Template MSA you authored
- Mined from your verified obligation history (n ≥ 3)
- Always tagged for review — never silently authoritative
Authority: Tier 2. Surfaces in the playbook with a Verify button.
What the market commonly does — when you are silent.
- 200+ curated UK / EU / US positions
- Scoped to MSA / NDA / SOW / OrderForm / DPA / Reseller / Licence
- Used as Tier 3 fallback when User and AI are both silent
Authority: Lowest. Citable in negotiations; opt-out per tenant.
Scoped resolution
One scope, one resolved playbook.
Vendor-side MSA, UK governing law. Every obligation type returns one resolved position with the tier that produced it.
| Type | Resolved position | Source |
|---|---|---|
| Liability | 100% fees in prior 12 months; uncapped for confidentiality, IP, mandatory law | User · Entered |
| Confidentiality | 3-year tail; trade secrets indefinite | User · Verified |
| Payment | 30 days net of valid invoice | AI · Template |
| Indemnification | IPR indemnity with mitigation; combination & modification carve-outs | AI · Contract |
| Notice | 30 days written, by either party | Market Practice |
| Renewal | Fixed term; no auto-renew; 60-day non-renewal window | Market Practice |
Tier 2 rows surface with a Verify button — click to promote into User. Tier 3 is the Market Practice corpus; opt out per tenant if your house view is silent for a reason.
The roster
12 specialists behind the platform.
Each capability is a scoped, named, individually priced agent. Extractor reads the PDF; Verifier holds it until a human signs off; Deviations compares each clause to your scoped playbook; Negotiator drafts the counter from your clause library; Renewals assembles the pack; Ask CLM answers portfolio questions in natural language. Mix the ones you need.
- ExtractorAgent
- PDF / DOCX / MD / TXT support
- Tool-use extraction with guaranteed schema adherence
- Per-obligation citation: verbatim quote + page + bbox
Extractor
Clause + obligation parserTurns every contract — PDF, Word, Markdown — into structured obligations.
From signed PDF to ~80 verifiable obligations per contract, each citation-pinned to the source paragraph and bounding-box.
From £600/moDetails - VerifierAgent
- Three-state verification: AI_Extracted → Verified / Overridden
- Review queue with value × age priority
- Per-binding ERP/CRM gate — unverified rows excluded
Verifier
Human-in-the-loop reviewerRoutes every AI-extracted obligation to a human before it binds anything.
Zero downstream connector (ERP/CRM/Procurement) trusts an unverified obligation. Review queue prioritises by value × age × confidence.
From £400/moDetails - DeviationsAgent
- Tier-aware comparison (Tier 1 / Tier 2 / Manual)
- Risk-tiered findings (High / Medium / Low)
- Teams adaptive-card escalation for High
Deviations
Standard-vs-actual auditorFlags every contract clause that departs from your approved standards.
Risk-tiered (High / Medium / Low) per clause, with the standard and the actual side-by-side. High-risk auto-routes to General Counsel via Teams adaptive card.
From £700/moDetails - NegotiatorAgent
- Per-deviation negotiation pack
- Suggested redline (Markdown, copyable)
- Sample email response paragraph
Negotiator
Counter-position drafterGenerates redlines anchored to your clause library — never invented prose.
For any flagged deviation: impact, counter-position, rationale, suggested redline (Markdown), and a sample email response — all grounded in clauses you've already approved.
From £950/moDetails - DrafterAgent
- 5 draft types: renewal / non-renewal / amendment / termination / brief
- Grounded in counterparty consolidated view
- deviations_from_standard JSON block on every draft
Drafter
Renewal / amendment / termination drafterDrafts the document — informed by every prior amendment with this counterparty.
Renewal confirmation, non-renewal notice, amendment proposal, termination notice, or internal briefing — all consolidated from the counterparty's contract history with [TO CONFIRM] placeholders instead of hallucinations.
From £800/moDetails - RenewalsAgent
- Upcoming renewals list with days-to-expiry urgency
- Per-contract renewal pack with recommended action
- Standard-vs-actual table with tier provenance
Renewals
Lifecycle orchestratorBuilds a one-shot renewal pack 60–120 days before every expiry.
Recommended action (Renew / Renegotiate / Confirm or notify non-renewal / Review) + standard-vs-actual table + ready-to-send draft — surfaced in time to act.
From £550/moDetails - RiskAgent
- Composite 0–100 score with reason breakdown
- Bands: Low / Medium / High
- Portfolio-wide ranking for triage
Risk
Counterparty risk scorerComposite 0–100 score across deviations, exposure, verification ratio, and age.
Sorts your portfolio by attention-needed. Risk band (Low / Medium / High) with reason breakdown — no opaque scores.
From £450/moDetails - Standards LearnerAgent
- Auto-derives Tier 1 from SharePoint /Templates
- Mines Tier 2 from verified historical obligations
- Tier provenance on every active standard
Standards Learner
Tiered standards keeperDerives your negotiation baseline from your own template drafts and verified history.
Three-tier authority: Tier 1 (templates you authored) > Tier 2 (statistical learning from executed contracts) > Manual. The deviation engine and the drafter both consume the same resolved standard.
From £350/moDetails - WatcherAgent
- Per-tenant bound folders (Templates / Agreements roles)
- Cursor-based ingest — never misses, never re-ingests
- On-bind backfill of existing files
Watcher
SharePoint folder ingestBind a folder; new files become structured obligations within 5 minutes.
Zero Power Automate flows. Bind your existing SharePoint folders to roles (Templates / Agreements); a cron job + Graph delta query handles the rest.
IncludedDetails - Ask CLMAgent
- Natural-language Q&A over contracts + obligations
- Citation chips with hover preview + click-through to grid
- Persisted chat history per user, soft-delete + audit
Ask CLM
Conversational portfolio queryerNatural-language questions over your obligation graph, citation-grounded.
Type 'show me all payment obligations above £50k due in the next 90 days' — get an answer with clickable obligation chips that jump straight into the grid.
From £400/moDetails - CuratorAgent
- Per-type clause library with preference tiers
- Derive-from-templates button (Tier-1 seed)
- Admin CRUD via Clauses tab
Curator
Clause library keeperThe approved clause language your negotiators reuse.
Per-obligation-type library of preferred / fallback / walk-away clauses. Negotiator drafts redlines using this text verbatim — no inventing contract language.
From £250/moDetails - LifecycleAgent
- Explicit lifecycle event recording with audit trail
- 'Unverified — needs review' first-class posture
- 90 / 60 / 30 / 14 day alert tiers
Lifecycle
Posture + event recorderThe system flags, the user decides. Never silently auto-decides.
Every active / renewed / terminated / expired transition is a human-recorded event with an audit trail. 'Unverified — needs review' is a first-class posture state.
IncludedDetails
Pricing scales with the mix. À-la-carte from £250/mo.
See tiers, agent prices, and how we compare →Product tour
Three artefacts your team works from on day two.
Citation-grounded obligations. A coverage heatmap that shows where your playbook is silent. A deviation pack with a copyable counter-position. No prompt-engineering required — these are the surfaces that ship.
Every obligation, citation-grounded to the source PDF.
Claude Haiku does tool_use extraction into a strict schema. Each row carries a verbatim quote, page number, and a bounding box anchored to the source PDF — click it and the PDF viewer jumps to the exact clause. AI-extracted rows are held in the Review queue until a human verifies them; downstream connectors gate on that state.
Acme MSA 2024 · Obligations
81 extracted · 64 verified- ConfidentialityVerified
3-year survival of confidentiality post-termination
p.7· - LiabilityVerified
Capped at 12 months' fees; uncapped for confidentiality, IP
p.14· - RenewalUnverified
Auto-renew 12mo unless 60-day non-renewal notice
p.21·
“Each party’s aggregate liability arising under or in connection with this Agreement shall not exceed the fees paid in the twelve (12) months preceding the claim, save for breach of confidentiality, infringement of intellectual property, or liability which cannot be excluded by law.”
See where your playbook is silent, before counterparty paper finds it.
The coverage heatmap maps scope (agreement_type × party_role × jurisdiction) against obligation type. Green is your User-tier house view. Amber is AI-derived from your templates and history, awaiting verification. Sky is Market Practice fallback. Greys are gaps — write a position, or accept the market position with one click.
Playbook coverage
42 of 49 cells covered · 7 gaps| Scope | Liability | Confidentiality | Payment | Notice | IP | DP | Termination |
|---|---|---|---|---|---|---|---|
| MSA · vendor · UK | |||||||
| MSA · customer · UK | |||||||
| NDA · mutual | |||||||
| OrderForm · vendor | |||||||
| DPA · processor · EU | |||||||
| MSA · vendor · US-NY |
Flagged clauses come with a redline, not just a warning.
Every departure from the resolved scoped position opens into a negotiation pack: what they drafted, your house view, the suggested redline drawn verbatim from your clause library, and a sample paragraph for the reply email. Copy the redline; send the email; close the gap.
Liability · clause 11.2
High riskAcme Holdings · MSA · vendor · UK · flagged on extraction
“No cap on indirect or consequential losses for breaches of confidentiality, IP, or applicable law.”
Aggregate cap = 100% of fees paid in the prior 12 months; uncapped only for confidentiality, IP indemnity, and mandatory law.
Each party’s aggregate liability shall not exceed 100% of the fees paid in the twelve (12) months preceding the claim, save for breach of confidentiality, infringement of intellectual property, or liability which cannot be excluded by law…
Trust & architecture
What your security team will want in writing.
Legal-grade reliability is an architecture problem, not a prompt-engineering problem. CLMSpace enforces the load-bearing guarantees below in code, not in policy.
Multi-tenant isolation by architecture
Per-tenant Dataverse environment for the obligation graph; per-tenant configuration row (rc_tenantconfig) keyed by tenant id; per-tenant Azure AD app registration for Microsoft Graph + Dataverse. The API resolves a tenant id from the bearer key into a Python contextvar on every request and every downstream call inherits that scope. No shared business-data tables exist — a mis-scoped query fails closed rather than crossing tenants.
UK GDPR · EU GDPR · data residency pinned
The Azure Container App runtime and ACR run in UK South today. For EU residency-constrained customers, the Dataverse environment can be provisioned in a Microsoft EU data centre at onboarding; the API region follows. Residency is pinned at the data layer (per-tenant Dataverse environment) rather than at the application layer, so a tenant cannot accidentally write across regions. DPA, ROPA, and DPIA templates available on request.
Schrems II · IDTA-ready transfers
Where the inference call to Anthropic touches a US endpoint, the transfer is governed by the UK IDTA (or EU SCCs as appropriate) with a documented Transfer Impact Assessment and Anthropic's zero-retention enterprise contract as the supplementary technical measure. Your contract content is not retained beyond the request lifecycle and is not used to train models.
Citation-grounded, not just AI-generated
Every extracted obligation carries a verbatim source quote, a page number, and a bounding-box coordinate from the source PDF. The portal's citation popover highlights the exact sentence on the source page on hover. If the model cannot cite a passage verbatim, the obligation is not written — citation grounding is a correctness control, not just a UX feature.
Human-in-the-loop verification gate
AI-extracted obligations and AI-suggested standards land as Tier 2 — excluded from your authoritative house view and from downstream drafting, deviation, and counter-position generators until a user verifies them. Every verification is timestamped and attributed (rc_obligationedit, rc_lifecycleevent). CLMSpace ships draft intelligence for your legal team to act on. It is not legal advice and we say so on every deliverable.
Bind your existing SharePoint
No Power Automate flows to maintain. No parallel document store. Bind the SharePoint folders you already use in Settings; a poller queries Microsoft Graph delta tokens every 5 minutes and ingests new files. App-only token with Sites.Read.All + Files.Read.All, scoped to the folders you've bound — nothing else in your tenant is enumerated or read.
FAQ
Questions we hear on every first call.
How is this different from Ironclad, Icertis, or DocuSign CLM?
Traditional CLMs are forms-and-workflows tools. They store contracts, route signatures, and surface dates — but they still need a human to read each clause and remember each renewal. CLMSpace produces the same surface area (obligation register, clause library, renewal workflow) but every clause has been read by Claude, anchored to the source PDF, resolved against a scoped playbook, and surfaced as a verifiable obligation row before anyone opens the agreement. The differentiator is the scoped User → AI → Market Practice playbook and the verification gate that sits between AI output and downstream connectors — neither of which exists in the incumbent stack.
Is CLMSpace a GRC tool? Compliance? Audit?
No — and this matters. CLMSpace is a contract-lifecycle platform: extraction, playbook, drafting, deviation, renewal. It does not run controls libraries, regulatory horizon scans, evidence catalogues, internal audit workflows, or risk registers. If you need GRC, pair us with a GRC tool — they operate on different objects (controls vs. obligations) and answer different questions (am I compliant with X regulation vs. what does my MSA with Acme actually say). We will gladly help your security team write a clear scoping note for procurement.
How does the playbook authority model work?
Three tiers, resolved in order. Tier 1 (User) is anything your admin entered, imported, or verified — the highest authority, quoted verbatim in drafts. Tier 2 (AI) is what we derived from the templates you bound and from your verified obligation history (statistical, minimum sample size of 3) — surfaces with a 'Verify' button to promote into Tier 1. Tier 3 (Market Practice) is a curated 200+ entry corpus of common UK / EU / US positions we ship with the platform — used as fallback when User and AI are silent for that scope. Every position is scoped by agreement_type × party_role × jurisdiction, so the resolver returns different defaults for vendor-side MSA, buyer-side OrderForm, US-NY MSA, EU DPA, etc.
Is this legal advice?
No, and we say so on every deliverable. CLMSpace produces draft contract intelligence for your qualified counsel to act on. We do not form a lawyer-client relationship and we do not displace your legal team's judgement. The verification gate exists so a human reviews every AI-extracted obligation before it can affect a downstream system or a draft.
What if an agent hallucinates an obligation or a clause?
It cannot reach a downstream connector or a published draft. Every AI-extracted obligation lands in the Review queue with state AI_Extracted; the ERP payment schedule, CRM counterparty snapshot, and Procurement spend-cap exports all gate on Human_Verified (or Human_Overridden) state by default. For drafting and counter-positions, the Negotiator agent is constrained to reuse approved clause-library text verbatim — it does not invent contractual language. Extraction itself uses Claude tool_use with a strict JSON schema, so the model cannot return prose that wasn't in the source PDF.
Do you train on our contracts?
No. We use zero-retention inference endpoints where our model providers support them, and we do not fine-tune or train on any client data. Your contracts live in your Microsoft 365 tenant; CLMSpace reads them via Microsoft Graph with app-scoped permissions, extracts structured obligations into your per-tenant Dataverse environment, and never copies content into a CLMSpace-owned store.
Does this work with our existing SharePoint / Microsoft 365 stack?
That is the design point. Bind the SharePoint folders you already keep contracts in; the Watcher polls Microsoft Graph delta queries every 5 minutes. No Power Automate flows to maintain. No parallel document store. Single sign-on through Entra. The Teams app surfaces alerts and approval cards in the channel your team already lives in.
Which contract types and file formats do you cover?
MSAs, NDAs, SOWs, OrderForms, online terms, variations, amendments, DPAs, reseller and licence agreements, and bespoke counterparty paper. File formats: .pdf (with PyMuPDF word-level bbox), .docx (paragraph + table parsing), .md, .txt. PDF citations are click-through with bounding boxes; non-PDF formats use char-offset citations.
Can I start with one tier and grow into more agents later?
Yes. Solo (Extractor + Verifier + Lifecycle) is the entry point for cleaning up the contract register. Adding Drafter, Renewals, Deviations, Watcher, and Ask CLM (Foundation) is a billing-day upgrade with no migration. Negotiator, Risk, Standards Learner, and Curator move you to Commercial Office. Cancel any agent for the next billing period; you stop paying for it.
What is the deployment time?
One day for Solo — bind a SharePoint folder; ingestion runs in the background within 5 minutes. One week for Foundation — template uploads feed Tier-1 positions, Renewals starts assembling packs, Ask CLM answers portfolio questions. Two weeks for Commercial Office — clause library curation, deviation thresholds, counter-position policy review. The first renewal pack typically lands within a week of binding the Agreements folder.
See it run on your contracts.
Bind a SharePoint folder; the Watcher ingests within 5 minutes. Solo runs the same day. Foundation is up the same week: Tier-1 positions seeded from your template uploads, the first renewal pack drafted, Ask CLM answering portfolio questions on day two. Commercial Office adds clause-library curation and the scoped playbook over the following week.