Synth AI SDK

Typed Index v0.2 APIs include QA and lifecycle operations, exact-revision rights attestations, and resealed research corrections.

Python SDK, CLI and MCP servers for Synth Index (cited search over reviewed research) and Swarms (hosted research runs under client.research.swarms).

Documentation: https://docs.usesynth.ai/sdk/overview

Installation

uv add synth-ai

Authenticate

Free anonymous public Index Search needs no key (see Synth Index). Keyed Index Search, private collections and Swarms need SYNTH_API_KEY:

export SYNTH_API_KEY="sk_..."

A key selects the paid keyed Index route; it never makes a public request free.

Local Workspaces

For local multi-repo development, synth-ai treats workspace resolution as a read-only overlay. .env and Synth home config may provide defaults and secrets; selecting a worktree must not rewrite those defaults.

Use SYNTH_WORKSPACE_MANIFEST or SYNTH_WORKSPACE_ROOT for command-scoped worktree resolution. The resolver in synth_ai.core.utils.workspace returns repo paths and a scoped env mapping for subprocesses without mutating .env.

Pass base_url when you need to pin a production, local, staging, or private backend explicitly:

from synth_ai import SynthClient

client = SynthClient(base_url="http://127.0.0.1:8000")

The CLI also reads SYNTH_BACKEND_URL and accepts --backend-url.

Quickstart: Swarms

from synth_ai import SynthClient
from synth_ai.sdk.research.public import SwarmSpec

with SynthClient() as client:
    swarm = client.research.swarms.create(
        SwarmSpec(objective="Assess this repository and produce a concise report.")
    )
    for event in swarm.events():
        print(event.kind, event.telemetry.sequence)
    result = swarm.wait(timeout_seconds=900)
    print(result.swarm_id, result.state)
    resolved = swarm.configuration()
    print(resolved.config_version_id, resolved.snapshot_sha256)
    usage = swarm.usage()
    print(usage.money.nominal_pico_usd, usage.tokens.totals.input_tokens)
    evidence = swarm.evidence()
    print(evidence.artifacts, evidence.work_products)

create returns a durable handle immediately. events() yields typed events, including an explicit UnknownSwarmEvent for forward-compatible server events; wait() uses a monotonic deadline and returns the terminal typed Swarm. configuration() returns the immutable, versioned, secret-redacted launch snapshot bound to that swarm, so replay and audit do not depend on the project's current mutable configuration. usage() returns one typed cost, token, and actor-attribution projection plus its source, record count, observation time, and terminal-state freshness. It does not expose the legacy raw ledger-entry dictionaries. evidence() returns the complete durable artifact and WorkProduct index with strict counts and lifecycle freshness. Artifact and WorkProduct content reads use the same typed transport and return bytes; they do not expose storage authority.

Research SDK (Swarms)

SynthClient().research hosts Swarms. Its stable namespaces are projects, swarms, and factories (the last is a supported compatibility API; it is not part of this release's recommended path).

Create a durable project when work needs reusable configuration:

from synth_ai import SynthClient
from synth_ai.sdk.research.public import EnvironmentKind, ProjectSpec, RuntimeKind, SwarmSpec

with SynthClient() as client:
    project = client.research.projects.create(
        ProjectSpec(
            name="Repository assessment",
            pool_id="pool_default",
            runtime_kind=RuntimeKind("python"),
            environment_kind=EnvironmentKind("docker"),
            orchestrator_profile_id="profile_orchestrator",
            default_worker_profile_id="profile_worker",
        )
    )
    swarm = client.research.swarms.create(
        SwarmSpec(objective="Produce the assessment."),
        project_id=project.project_id,
    )
    print(swarm.wait().state)

Limits, economics, secrets, Tag, rich evidence projections, and administrative resource APIs remain available under client.research.advanced while their contracts are stabilized. Advanced APIs are not covered by the stable surface guarantee.

CLI discovery:

synth-ai research --help

Project creation also accepts the backend-owned ProjectSpec.policy mapping. For example, a server-enabled fresh project can request policy={"host_resource_custody_mode": "horizons_docker_sessions_only"}. The backend validates this restricted mode and owns its immutable resource binding; SDK serialization does not grant additional authority.

Container pools

The optional synth-ai[pools] extra exposes the canonical synth-containers client through AsyncSynthClient.pools. It uses the same configured backend credential and keeps hosted admission, resource ownership, and recovery in the backend. The enclosing async client closes the pool transport.

from synth_ai import AsyncSynthClient

async def inspect_lease(lease_id: str, task_id: str):
    async with AsyncSynthClient() as client:
        return await client.pools.get_lease_interactive(lease_id, task_id=task_id)

For explicit lifetime management, from synth_ai.pools import PoolClient re-exports the same implementation. Research-only installations do not import this optional dependency. Development candidates must install the exact pinned containers wheel; an unpublished candidate extra is not a release claim.

CLI

synth-ai --help
synth-ai index --help      # Synth Index Search and research intake
synth-ai research --help   # Swarms and Research projects

Synth Index

Index Search returns a cited response plus the exact Contribution revisions it cites. There are two routes, chosen explicitly by the client you use:

	Public route	Keyed route
Caller	Anonymous only: no API key or session is sent	SYNTH_API_KEY (or a Clerk session)
Python	PublicIndexClient().public_search(...)	SynthClient().index.search(...)
CLI	synth-ai index search --public (0.21.1+)	synth-ai index search --keyed
MCP index_search	No SYNTH_API_KEY in the server env	SYNTH_API_KEY set; runs only after the org turns on wallet payments
Corpus	Published public Contributions	Public, or authorized private collections
Price	Free to the caller	Paid: Fast is 5 cents per delivered Search; Deep is 5 cents plus measured model cost, capped by your max_charge_cents and at $1 per Search. The backend publishes the current price.
Funding	None	Index promo credit, then the wallet with explicit consent (allow_wallet, max_charge_cents)
Limits	Per-caller and platform-wide per-minute/per-day limits, a daily service budget and a Deep concurrency cap, published in capabilities().public_search	Your organization's mode grants, monthly caps and per-mode concurrency
Retention	Published in capabilities().public_search (retention, privacy_copy)	Your organization's terms; public terms do not describe keyed Search

The public route exists only where the backend enables it (public_search.enabled in capabilities); otherwise it fails closed with PublicSearchDisabledError. The public route refuses any credential (409, PublicSearchAuthenticatedError), so a key never turns a public request into a free keyed one, and the SDK and CLI never switch routes on their own.

from synth_ai.sdk.index import PublicIndexClient

with PublicIndexClient() as index:  # anonymous: sends no key
    print(index.public_search_terms().lines)  # price, limits, privacy from capabilities
    result = index.public_search("RLVR verifier design", mode="fast", max_results=5)
    print(result.response, [item.citation for item in result.citations])

Keyed (paid) Search:

from uuid import uuid4

from synth_ai import SynthClient
from synth_ai.sdk.index import SearchBillingConstraints

request_key = str(uuid4())  # Persist this before sending; reuse it on uncertain retry.
with SynthClient() as synth:  # Reads SYNTH_API_KEY.
    result = synth.index.search(
        query="What evidence supports the retrieval design?",
        mode="fast",
        billing=SearchBillingConstraints(allow_wallet=True, max_charge_cents=5),
        idempotency_key=request_key,
    )
    print(result.response, result.usage)

CLI:

synth-ai index search "RLVR verifier design" --public
synth-ai index search "RLVR verifier design" --keyed --allow-wallet --max-charge-cents 5

With no key and no route flag the CLI refuses and asks for one. With an inherited SYNTH_API_KEY, --public still sends no key; without a flag the CLI uses the keyed route and says so on stderr.

• MCP: synth-ai-index-mcp is an Index-only stdio server, read-only by default. Setup and per-configuration economics are in the Index SDK guide.
• Errors: public-route refusals have their own types (rate limit, budget exhausted, disabled route, credential conflict); see the Index SDK guide.

These calls require a deployed Index API; installing the SDK alone does not make a Search available.

Public Surface

Use SynthClient as the front door:

Surface	Client namespace	Use it for
Index	client.index, PublicIndexClient	Keyed (paid) and anonymous public (free) Search, exact contents.
Swarms	client.research.swarms	Hosted research runs, events, usage and evidence.
Research projects	client.research.projects	Reusable Swarm configuration.
CLI / MCP	synth-ai, synth-ai-index-mcp	Terminal commands and an Index-only coding-agent server.

Swarms run on Synth's hosted research workers; see the Managed Research docs for repo runs, evidence, checkpoints and final reports.

Swarms Billing

Swarms and other Managed Research work draw from the same org-level allowance and flex-credit wallet. Free, Standard ($20/month), and Max ($200/month) expose premium and value usage windows with reset times, then use explicit flex credits after included usage is exhausted. Premium models consume allowance faster; value models stretch the same allowance further. Promo, make-good, banked, and override grants are manual audit events rather than automatic resets.

The canonical backend surfaces are GET /smr/billing/catalog, GET /smr/billing/plan, GET /smr/billing/runs/{run_id}/drawdown, and GET /smr/billing/factory-efforts/{factory_effort_id}/drawdown. In the Python SDK, use client.research.advanced.economics for authoritative billing reads while the economics contract remains advanced. Do not infer allowance from legacy Autumn balances or local spend summaries, and do not recompute discounts in the client.

Links

• Install and authenticate
• SynthClient guide
• SDK reference
• OpenAPI contracts

Local Development

Use uv run for Python tools:

uv sync --group dev
uv run ruff format --check .
uv run ruff check .
uv run ty check
make docs-gen   # generate Blume SDK reference into docs/
make docs-dev   # preview at http://localhost:3000/overview

Optional: install Lefthook and run lefthook install to run formatting, linting, and type checks on staged Python files.