Ishara.

For AI builders

Business context for your models and agents,
under your access rules

Give your agents the people, companies, relationships and records your team can see, through the Context API and Object API with scoped keys. Agents propose changes and a person approves them. An MCP server is available on request.

A work email reserves $50 in credit.

Context API and Object API · Scoped keys · MCP server on request

What an agent can read

Ask the Context API for context, not rows

One call returns what Ishara knows about a person, a company or a meeting: who they are, how your team knows them, what was said and what is still open. Your agent reads it under the key owner's access, and nothing it reads is used to train a model.

AskRoute (and MCP tool)What comes back
Look up a personPOST /v1/people/contextishara_people_contextWho they are, their position and company, and how your team knows them.
Look up a company and its teamPOST /v1/companies/contextPOST /v1/companies/teamishara_companies_contextishara_company_teamThe company's facts, how its team divides, and the people on it.
Context for a meeting or an accountPOST /v1/context/bundleishara_context_bundlePeople, companies, history, open follow-ups, email subjects and deals, in one call.
Relationship historyPOST /v1/relationships/historyishara_relationship_historyEvery earlier meeting with these people, and when the first conversation was.
Pre-meeting briefPOST /v1/briefs/getishara_brief_getThe brief Ishara prepared for a meeting, with its document and delivery record.
Open follow-upsPOST /v1/followups/openishara_followups_openTasks and promises still open with these records, with where each came from.

One request, one answer your model can use

Name the people and the sections you want. Ishara resolves who they are, reads each section as the person the key belongs to, and returns one document with where every fact came from and what was left out, and why.

  • Email context carries subjects, direction and who wrote last. It never carries message bodies.
  • Every call is logged with its purpose, so agent reads are told apart from people's.
  • An administrator can switch context sharing off for the whole workspace.

The full OpenAPI reference comes with your API access.

Example requestcurl
curl -X POST "https://<your-ishara-api-host>/v1/context/bundle" \
  -H "X-API-Key: $ISHARA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "attendees": [
      { "email": "layla@example.com", "name": "Layla Haddad" }
    ],
    "sections": ["people", "history", "followups", "mail"]
  }'

What an agent can change

An agent suggests the change, and a person applies it

Through the API's proposals route, an agent sends a change with its reason and evidence. Ishara checks it against your rules and shows the record before and after. Nothing on the record moves until a person applies it in Review.

  1. ProposeWith a reason and evidence
  2. PreviewBefore and after, per record
  3. A person appliesOr rejects, in Review
  4. On the recordMarked as AI, with its run

Low-risk proposals, such as a note or a task, apply themselves only in a workspace whose administrator switched that on. It is off by default. More on AI you approve before it writes anything.

Object API

Build on your records without a second copy

The same routes serve Companies, People, Deals and every object you define. One key in the X-API-Key header. Writes are idempotent under an Idempotency-Key, so a retried request writes once; an upsert matches on any identity key you name; a batch sends many operations in one request.

MethodPathDoes
GET/objectsEvery object with its attributes, so a client describes itself
GET/objects/:object/recordsList records of any object, with search, filters and sort
POST/objects/:object/recordsCreate a record, validated against its field types
POST/objects/:object/records/upsertCreate or update, matched on any identity key you name
GET/records/:idOne record with its values and list entries
PATCH/records/:idUpdate values
DELETE/records/:idArchive a record; it stays restorable
GET/searchSearch across your objects
GET/lists/:id/entriesThe records on a list, with their list values
POST/records/:id/notesAdd a note; tasks and logged interactions work the same way
POST/batchMany operations in one request, optionally all or nothing
POST/proposalsPropose changes for a person to approve
GET/eventsPull events by cursor, for systems that prefer to ask
POST/subscriptionsSend events to your endpoint, signed

Keys are created in Settings and shown once. Each has scopes per resource (read companies, write notes), a rate limit, an expiry by default and overlapping rotation. An integration gets its own identity, so its key outlives whoever issued it. Calls run as the key's owner, so validation and access rules apply exactly as they do in the app.

Webhooks

Every change reaches your systems, signed

A record changes and every subscribed endpoint hears about it: signed with HMAC-SHA256 over the body and a timestamp, retried with growing gaps, replayable from the delivery log, and available to pull by cursor.

  1. EventA record changes
  2. Fan-outOne delivery per endpoint
  3. RetriesOnly if a delivery fails
  4. DeliveredSigned and timestamped
  5. ReplayFrom the delivery log

Secrets are shown once and rotate at any time. Endpoints are https only. A webhook is also one action an automation can take, and the integrations page shows what connects without code.

How you connect

REST with a scoped key, and MCP when you ask

Anything that can send an HTTPS request can call the Context API and the Object API: your own agent, an integration or a script. Each one uses a key your workspace issues, with only the scopes it needs.

  • Give a key context:read to read, and proposals:propose to suggest changes.
  • Only a key that a member deliberately grants proposals:apply can change a record.
  • Revoke the key in Settings and its next call is refused.

For assistants and tools that speak MCP, an Ishara MCP server is available on request. It wraps the same read routes, one tool per route, holds only the key you give it, and every tool call is one request to the API, so the same scopes, limits and request log apply. Ask for MCP access, with what you are building.

  1. Get early access

    Invites go out in waves. A work email reserves $50 in credit for your workspace.

  2. Create a key

    In Settings, under Integrations, choose what it may do: read context, read records, propose changes. It acts as you, under your access.

  3. Call the API

    From your agent or your system. Any change it proposes waits in Review for a person.

Plans and allowances

What each plan gives your agents

Prices exclude VAT. Early-access prices are kept for 12 months for workspaces that pay before 15 March 2027, and Pro and Business are arranged with our team.

PlanPriceAPI allowanceWhat it adds
Free$01,000 reads a month, read onlyRead-only access, to try it on your own records
Pro$12 a seat a month, or $10 billed yearly50,000 calls a monthAPI keys with scopes, webhooks and automations
Business$39 a seat a month, or $32.50 billed yearly500,000 calls a monthEverything in Pro, at team scale
Bid-team edition$7,500 a year, per organisation1,000,000 calls a monthBusiness for a whole bid team, scoped with our team
EnterpriseScoped with your teamScopedAgent access at organisation scale

Under your rules

The API is the boundary

A key acts as its owner

It sees the records, threads and calls that person can see, nothing more, narrowed further by its scopes.

Every read is recorded

Each Context API call is logged with its purpose. Each Object API call is logged too.

Never used for training

Your private context is supplied at the moment of a request and is never used to train models.

You hold the switch

Revoke a key, rotate it, or switch context sharing off for the workspace. Security details are on the security page.

Frequently asked questions

Is our data used to train AI models?

No. Your private context is supplied to a model at the moment of a request, under the key owner's access, and is never used to train models.

Which agents can connect to Ishara?

Any system that can send an HTTPS request can call the API, and an MCP server is available on request. Each connection uses a key your workspace issues, with the scopes you choose.

Does an agent read our email?

Through the Context API it reads who wrote, when and the subject of each thread on the records it asks about. It never receives message bodies.

Can an agent change our records?

It proposes. A person applies the proposal in Review, unless a member deliberately gives a key the apply scope. An integration key you give write scopes writes through the Object API as its owner, under the same access rules.

Which plan includes the API?

Free includes read-only access, 1,000 reads a month. API keys, webhooks and 50,000 calls a month come with Pro at $12 per seat per month ($10 billed yearly), and Business carries 500,000. Ishara opens in waves from the waiting list, and a work email reserves $50 in credit.

How do we get the MCP server?

Ask for it: an MCP server is available on request. Email our team with what you are building, and it runs under the same scoped key and access rules as the API.

Can we switch agent access off?

Yes. Revoke or rotate any key in Settings, and an administrator can switch context sharing off for the whole workspace.

Give your agents context they can trust

Get early access with a work email to reserve $50 in credit, or email our team about your use case, MCP included.

Invite-only early access in waves · A work email reserves $50 in credit · Arabic and English