silas 0.2.0 → 0.3.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/CHANGELOG.md +61 -1
- data/README.md +44 -0
- data/app/controllers/concerns/silas/api/serialization.rb +63 -0
- data/app/controllers/silas/api/base_controller.rb +24 -0
- data/app/controllers/silas/api/v1/approvals_controller.rb +35 -0
- data/app/controllers/silas/api/v1/sessions_controller.rb +39 -0
- data/app/controllers/silas/api/v1/streams_controller.rb +117 -0
- data/app/controllers/silas/api/v1/turns_controller.rb +32 -0
- data/app/controllers/silas/inbox/sessions_controller.rb +16 -3
- data/app/controllers/silas/inbox/turns_controller.rb +19 -0
- data/app/helpers/silas/inbox/trace_helper.rb +5 -0
- data/app/models/silas/session.rb +6 -0
- data/app/models/silas/turn.rb +9 -0
- data/app/views/layouts/silas/inbox.html.erb +6 -0
- data/app/views/silas/inbox/sessions/index.html.erb +11 -3
- data/app/views/silas/inbox/steps/_step.html.erb +3 -0
- data/app/views/silas/inbox/turns/_header.html.erb +12 -1
- data/config/routes.rb +23 -1
- data/db/migrate/20260725000001_add_provider_to_silas_steps.rb +13 -0
- data/lib/generators/silas/install/install_generator.rb +1 -0
- data/lib/generators/silas/install/templates/agent.yml +10 -1
- data/lib/generators/silas/install/templates/initializer.rb +5 -4
- data/lib/silas/agent.rb +4 -0
- data/lib/silas/chat.rb +2 -0
- data/lib/silas/configuration.rb +30 -34
- data/lib/silas/doctor.rb +155 -0
- data/lib/silas/engines/ruby_llm.rb +13 -2
- data/lib/silas/eval/assertions.rb +18 -0
- data/lib/silas/eval/transcript.rb +1 -0
- data/lib/silas/inbox/cost.rb +37 -13
- data/lib/silas/message_builder.rb +9 -2
- data/lib/silas/named_agent.rb +1 -0
- data/lib/silas/registry.rb +18 -8
- data/lib/silas/step_runner.rb +12 -0
- data/lib/silas/version.rb +1 -1
- data/lib/silas.rb +1 -0
- data/lib/tasks/silas_doctor.rake +17 -0
- metadata +10 -1
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 187a0730a1a84463f6207060d446c6960946513c0884d708fc0d16c011493a26
|
|
4
|
+
data.tar.gz: ed9a09b617c7dde80deaf6fe2c460e254cc96487b43eac52f56c6aa7dedecbaa
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: a38278a7020a9282b1ff9aaac7b18da60bd017a001bcc421c6e316d0033b1a4b468cd86c0d4bcdcd869940f7d9c4b953318e84dd626087bef4fee1ee8ea8b87c
|
|
7
|
+
data.tar.gz: 6f42e1a81bf935ebdbb39c678e46662ab85b9f45bd8293c0e83f4467e7db922e2102087ab62ea7eb8e8b1b3b1d323eba1a5772e997028b3e5838882bda9cfeb3
|
data/CHANGELOG.md
CHANGED
|
@@ -1,6 +1,66 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
-
## 0.
|
|
3
|
+
## 0.3.0
|
|
4
|
+
|
|
5
|
+
- **`bin/rails silas:doctor`** — every known first-run failure mode as one
|
|
6
|
+
command: provider key, queue adapter (async = red), model resolution with
|
|
7
|
+
prices, migrations, tool validation, the rescuer recurring entry, the cable
|
|
8
|
+
adapter for live streaming (async cable can't carry worker deltas), and
|
|
9
|
+
whether the deny-by-default inbox/API auth has been wired. Exits non-zero
|
|
10
|
+
on failures, so it slots into CI.
|
|
11
|
+
- **Hard-removed the 0.2 `:agent_sdk` config shims** — `config.auth` and
|
|
12
|
+
`config.agent_sdk_*` now raise `NoMethodError` (they were warning no-ops
|
|
13
|
+
for the 0.2 cycle). `config.engine = :agent_sdk` still raises the
|
|
14
|
+
explanatory `BootGuardError`. Workflows now use `actions/checkout@v5`
|
|
15
|
+
(Node 20 deprecation).
|
|
16
|
+
- **Inbox at scale + the promised top-up card.** Budget-parked turns now
|
|
17
|
+
render a **"Budget reached — raise & resume"** card (the UI the 0.1.5
|
|
18
|
+
changelog promised): one field, one click, `raise_budget!`, completed work
|
|
19
|
+
replays from rows. The session list pages by keyset (`?before=<id>`, 50 per
|
|
20
|
+
page — the old hard `limit(100)` made session 101 unreachable forever), and
|
|
21
|
+
the index renders in **2 queries** instead of ~4 per card.
|
|
22
|
+
- **Cost accounting prices itself from RubyLLM's model registry** (1,100+
|
|
23
|
+
models, refreshed upstream from models.dev) instead of a five-entry
|
|
24
|
+
hand-maintained map. Steps stamp the **provider** RubyLLM resolves at
|
|
25
|
+
persist time, and lookups use the two-arg `find(model, provider)` — 85/1081
|
|
26
|
+
registry ids exist under multiple providers at different prices, so the
|
|
27
|
+
bare lookup could silently price the wrong one. `config.model_prices` is
|
|
28
|
+
now an **override map** (default `{}`) for fine-tunes, custom deployments,
|
|
29
|
+
and models newer than the installed registry; unknown models stay
|
|
30
|
+
`unpriced`, never a lying $0.00. Dropped the write-only
|
|
31
|
+
`silas_turns.cost_microcents` column (declared since 0.1.0, never
|
|
32
|
+
populated). Default model is now `claude-sonnet-4-5` — in every supported
|
|
33
|
+
registry, unlike `claude-sonnet-5` (absent from ruby_llm 1.16's shipped
|
|
34
|
+
registry, which would have broken first runs), and never the priciest
|
|
35
|
+
model. New migration: `bin/rails silas:install:migrations db:migrate`.
|
|
36
|
+
- **Structured final answers.** Declare a JSON schema under `final_answer:` in
|
|
37
|
+
agent.yml and the turn's answer comes back as a parsed Hash —
|
|
38
|
+
`Turn#answer_data` (alongside `answer_text` for prose agents), in the API
|
|
39
|
+
(`answer_data` on turns, `structured` on steps), the inbox trace, the REPL,
|
|
40
|
+
and a new `assert_answer_data` eval assertion. Implemented on RubyLLM's
|
|
41
|
+
`with_schema` (each provider's native structured-output dialect). The schema
|
|
42
|
+
is model-visible state: it folds into the definitions digest **only when
|
|
43
|
+
present** — schema-less agents keep a byte-identical digest, so turns parked
|
|
44
|
+
across the upgrade never trip `NondeterminismError` (pinned by spec); a
|
|
45
|
+
mid-turn schema change fails loudly by design. Structured answers replay
|
|
46
|
+
into later turns' history as their JSON text (deterministic).
|
|
47
|
+
|
|
48
|
+
- **A public HTTP + SSE session API** at `/silas/api/v1` — sessions
|
|
49
|
+
(create/show, `?trace=1` for the full transcript), turns (create — **409**
|
|
50
|
+
while one is active — and cancel), and approvals (list / approve / decline,
|
|
51
|
+
the exact same `approve!`/`decline!` as every other surface, stamped with
|
|
52
|
+
`config.api_actor`). Deny-by-default via `config.api_auth`, the inbox's
|
|
53
|
+
contract. `GET .../sessions/:id/stream` is server-sent events at row
|
|
54
|
+
granularity (turn / completed-step / invocation changes) with
|
|
55
|
+
`Last-Event-ID` resume — at-least-once, epoch-ms watermark ids, `?poll=1`
|
|
56
|
+
for a curl-friendly backlog-and-close, self-closing after
|
|
57
|
+
`api_stream_max_duration` with a `timeout` reconnect event. The AR
|
|
58
|
+
connection is released between polls; per-token streaming remains the
|
|
59
|
+
Turbo/browser feature (deltas live in the worker process; the gem takes no
|
|
60
|
+
cross-process bus dependency). `Session` gains `parent_session` /
|
|
61
|
+
`child_sessions` associations, and session JSON carries the lineage.
|
|
62
|
+
|
|
63
|
+
## 0.2.0
|
|
4
64
|
|
|
5
65
|
- **Token streaming, end to end.** The engine seam's `&on_event` block — dead
|
|
6
66
|
code since 0.1.0 — is live: the `:ruby_llm` engine streams the model
|
data/README.md
CHANGED
|
@@ -33,6 +33,7 @@ app/agent/
|
|
|
33
33
|
bundle add silas
|
|
34
34
|
bin/rails generate silas:install
|
|
35
35
|
bin/rails db:migrate
|
|
36
|
+
bin/rails silas:doctor # key · queue adapter · model · migrations · tools · rescuer · cable · auth
|
|
36
37
|
```
|
|
37
38
|
|
|
38
39
|
```ruby
|
|
@@ -209,6 +210,49 @@ zero risk to the durability contract. Custom sinks subscribe to the
|
|
|
209
210
|
text: }`, where `text` is the accumulated string so far — filter by ids;
|
|
210
211
|
notifications are process-global).
|
|
211
212
|
|
|
213
|
+
## Structured answers
|
|
214
|
+
|
|
215
|
+
Give the turn's final answer a schema in agent.yml and read it back as a Hash:
|
|
216
|
+
|
|
217
|
+
```yaml
|
|
218
|
+
final_answer:
|
|
219
|
+
type: object
|
|
220
|
+
properties:
|
|
221
|
+
verdict: { type: string }
|
|
222
|
+
amount_pence: { type: integer }
|
|
223
|
+
required: [verdict]
|
|
224
|
+
```
|
|
225
|
+
|
|
226
|
+
`Turn#answer_data` returns the parsed payload (`answer_text` stays for prose
|
|
227
|
+
agents); the HTTP API carries it as `answer_data`, evals get
|
|
228
|
+
`assert_answer_data(key: :verdict, value: "approve")`. Rendered through
|
|
229
|
+
RubyLLM's `with_schema`, so each provider's native structured-output mode is
|
|
230
|
+
used. The schema is model-visible state — changing it mid-turn fails the turn
|
|
231
|
+
loudly rather than resuming into a different contract.
|
|
232
|
+
|
|
233
|
+
## The HTTP API
|
|
234
|
+
|
|
235
|
+
Everything the operator surface can do, over JSON — mounted with the engine at
|
|
236
|
+
`/silas/api/v1`, **deny-by-default** like the inbox (wire `config.api_auth`;
|
|
237
|
+
`config.api_actor` names the identity recorded on approvals):
|
|
238
|
+
|
|
239
|
+
```sh
|
|
240
|
+
curl -X POST .../silas/api/v1/sessions -d "input=Refund order 42, £12.50"
|
|
241
|
+
curl .../silas/api/v1/sessions/1?trace=1 # turns + steps + tool calls
|
|
242
|
+
curl .../silas/api/v1/sessions/1/approvals # what's parked
|
|
243
|
+
curl -X POST .../silas/api/v1/approvals/7/approve # the same approve! as the inbox
|
|
244
|
+
curl -X POST .../silas/api/v1/sessions/1/turns -d "input=Now email them" # 409 if busy
|
|
245
|
+
curl -X POST .../silas/api/v1/turns/9/cancel
|
|
246
|
+
curl -N .../silas/api/v1/sessions/1/stream # server-sent events
|
|
247
|
+
```
|
|
248
|
+
|
|
249
|
+
The stream is SSE at **row granularity** — turn / completed-step / invocation
|
|
250
|
+
changes, at-least-once with `Last-Event-ID` resume (ids are epoch-ms
|
|
251
|
+
watermarks; `?poll=1` returns the backlog and closes, curl-friendly; streams
|
|
252
|
+
close themselves after `api_stream_max_duration` and clients reconnect).
|
|
253
|
+
Per-token streaming is deliberately the browser/Turbo feature — deltas live in
|
|
254
|
+
the worker process, and the gem requires no cross-process bus.
|
|
255
|
+
|
|
212
256
|
## The inbox
|
|
213
257
|
|
|
214
258
|
Mount the engine (the generator does this) and a live inbox appears at
|
|
@@ -0,0 +1,63 @@
|
|
|
1
|
+
module Silas
|
|
2
|
+
module Api
|
|
3
|
+
# Minimal, dependency-free JSON shapes. Deliberately explicit allowlists —
|
|
4
|
+
# never as_json a whole row (instruction snapshots and internal columns
|
|
5
|
+
# don't belong on the wire).
|
|
6
|
+
module Serialization
|
|
7
|
+
def session_json(session, include_turns: true, trace: false)
|
|
8
|
+
base = {
|
|
9
|
+
id: session.id, agent_name: session.agent_name,
|
|
10
|
+
channel: session.channel, status: session.status,
|
|
11
|
+
parent_session_id: session.parent_session_id,
|
|
12
|
+
child_session_ids: session.child_sessions.ids,
|
|
13
|
+
cost: Silas::Inbox::Cost.for_session(session),
|
|
14
|
+
created_at: session.created_at, updated_at: session.updated_at
|
|
15
|
+
}
|
|
16
|
+
return base unless include_turns
|
|
17
|
+
|
|
18
|
+
base.merge(turns: session.turns.map { |t| turn_json(t, trace: trace) })
|
|
19
|
+
end
|
|
20
|
+
|
|
21
|
+
def turn_json(turn, trace: false)
|
|
22
|
+
base = {
|
|
23
|
+
id: turn.id, session_id: turn.session_id, index: turn.index,
|
|
24
|
+
status: turn.status, input: turn.input,
|
|
25
|
+
failure_reason: turn.failure_reason,
|
|
26
|
+
answer_text: (turn.answer_text if turn.completed?),
|
|
27
|
+
answer_data: (turn.answer_data if turn.completed?),
|
|
28
|
+
created_at: turn.created_at, updated_at: turn.updated_at
|
|
29
|
+
}
|
|
30
|
+
return base unless trace
|
|
31
|
+
|
|
32
|
+
base.merge(steps: turn.steps.map do |step|
|
|
33
|
+
step_json(step).merge(invocations: step.tool_invocations.map { |i| invocation_json(i) })
|
|
34
|
+
end)
|
|
35
|
+
end
|
|
36
|
+
|
|
37
|
+
def step_json(step)
|
|
38
|
+
{
|
|
39
|
+
id: step.id, turn_id: step.turn_id, index: step.index,
|
|
40
|
+
status: step.status, terminal: step.terminal,
|
|
41
|
+
model: step.model, provider: step.provider,
|
|
42
|
+
text: Array(step.response_blocks).select { |b| b["type"] == "text" }
|
|
43
|
+
.map { |b| b["text"] }.join.presence,
|
|
44
|
+
structured: Array(step.response_blocks).reverse
|
|
45
|
+
.find { |b| b["type"] == "structured" }&.dig("data"),
|
|
46
|
+
updated_at: step.updated_at
|
|
47
|
+
}
|
|
48
|
+
end
|
|
49
|
+
|
|
50
|
+
def invocation_json(invocation)
|
|
51
|
+
{
|
|
52
|
+
id: invocation.id, turn_id: invocation.turn_id,
|
|
53
|
+
tool_name: invocation.tool_name, status: invocation.status,
|
|
54
|
+
approval_state: invocation.approval_state,
|
|
55
|
+
arguments: invocation.arguments, result: invocation.result,
|
|
56
|
+
error: invocation.error, approved_by: invocation.approved_by,
|
|
57
|
+
decline_reason: invocation.decline_reason,
|
|
58
|
+
updated_at: invocation.updated_at
|
|
59
|
+
}
|
|
60
|
+
end
|
|
61
|
+
end
|
|
62
|
+
end
|
|
63
|
+
end
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
module Silas
|
|
2
|
+
module Api
|
|
3
|
+
# JSON API base: ActionController::API (no CSRF, no layouts, no cookies),
|
|
4
|
+
# deny-by-default auth with the inbox's contract — the host lambda DENIES
|
|
5
|
+
# by rendering (or head-ing) and PASSES by not rendering.
|
|
6
|
+
class BaseController < ActionController::API
|
|
7
|
+
before_action :authenticate!
|
|
8
|
+
|
|
9
|
+
rescue_from ActiveRecord::RecordNotFound do
|
|
10
|
+
render json: { error: "not found" }, status: :not_found
|
|
11
|
+
end
|
|
12
|
+
|
|
13
|
+
private
|
|
14
|
+
|
|
15
|
+
def authenticate!
|
|
16
|
+
Silas.config.api_auth.call(self)
|
|
17
|
+
end
|
|
18
|
+
|
|
19
|
+
def current_actor
|
|
20
|
+
Silas.config.api_actor.call(self)
|
|
21
|
+
end
|
|
22
|
+
end
|
|
23
|
+
end
|
|
24
|
+
end
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
module Silas
|
|
2
|
+
module Api
|
|
3
|
+
module V1
|
|
4
|
+
# The headline feature over HTTP: list what's parked, approve or decline
|
|
5
|
+
# it — the exact same approve!/decline! as the inbox, Slack, and email.
|
|
6
|
+
class ApprovalsController < Silas::Api::BaseController
|
|
7
|
+
include Silas::Api::Serialization
|
|
8
|
+
|
|
9
|
+
# GET /silas/api/v1/sessions/:session_id/approvals
|
|
10
|
+
def index
|
|
11
|
+
session = Silas::Session.find(params[:session_id])
|
|
12
|
+
render json: { approvals: session.pending_approvals.order(:id).map { |i| invocation_json(i) } }
|
|
13
|
+
end
|
|
14
|
+
|
|
15
|
+
# POST /silas/api/v1/approvals/:id/approve
|
|
16
|
+
def approve
|
|
17
|
+
invocation = Silas::ToolInvocation.find(params[:id])
|
|
18
|
+
invocation.approve!(by: current_actor)
|
|
19
|
+
render json: invocation_json(invocation.reload)
|
|
20
|
+
rescue Silas::Error => e
|
|
21
|
+
render json: { error: e.message }, status: :conflict
|
|
22
|
+
end
|
|
23
|
+
|
|
24
|
+
# POST /silas/api/v1/approvals/:id/decline { reason: (optional) }
|
|
25
|
+
def decline
|
|
26
|
+
invocation = Silas::ToolInvocation.find(params[:id])
|
|
27
|
+
invocation.decline!(reason: params[:reason].presence || "declined via api", by: current_actor)
|
|
28
|
+
render json: invocation_json(invocation.reload)
|
|
29
|
+
rescue Silas::Error => e
|
|
30
|
+
render json: { error: e.message }, status: :conflict
|
|
31
|
+
end
|
|
32
|
+
end
|
|
33
|
+
end
|
|
34
|
+
end
|
|
35
|
+
end
|
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
module Silas
|
|
2
|
+
module Api
|
|
3
|
+
module V1
|
|
4
|
+
class SessionsController < Silas::Api::BaseController
|
|
5
|
+
include Silas::Api::Serialization
|
|
6
|
+
|
|
7
|
+
# POST /silas/api/v1/sessions { input:, agent: (optional), metadata: (optional) }
|
|
8
|
+
# channel stays nil ("direct") — API consumers read state via GET/stream;
|
|
9
|
+
# a non-nil channel would enqueue pointless outbound delivery jobs.
|
|
10
|
+
def create
|
|
11
|
+
input = params[:input].to_s.strip
|
|
12
|
+
return render json: { error: "input is required" }, status: :unprocessable_entity if input.empty?
|
|
13
|
+
|
|
14
|
+
handle = params[:agent].present? ? Silas.agent(params[:agent]) : Silas.agent
|
|
15
|
+
session = handle.start(input: input, metadata: metadata_param)
|
|
16
|
+
render json: session_json(session.reload), status: :created
|
|
17
|
+
rescue Silas::TurnInProgressError => e
|
|
18
|
+
render json: { error: e.message }, status: :conflict
|
|
19
|
+
rescue Silas::Error => e
|
|
20
|
+
render json: { error: e.message }, status: :unprocessable_entity
|
|
21
|
+
end
|
|
22
|
+
|
|
23
|
+
# GET /silas/api/v1/sessions/:id — session + turns
|
|
24
|
+
# GET /silas/api/v1/sessions/:id?trace=1 — plus steps + invocations
|
|
25
|
+
def show
|
|
26
|
+
session = Silas::Session.includes(turns: { steps: :tool_invocations }).find(params[:id])
|
|
27
|
+
render json: session_json(session, trace: params[:trace].present?)
|
|
28
|
+
end
|
|
29
|
+
|
|
30
|
+
private
|
|
31
|
+
|
|
32
|
+
def metadata_param
|
|
33
|
+
raw = params[:metadata]
|
|
34
|
+
raw.respond_to?(:to_unsafe_h) ? raw.to_unsafe_h : (raw.presence || {})
|
|
35
|
+
end
|
|
36
|
+
end
|
|
37
|
+
end
|
|
38
|
+
end
|
|
39
|
+
end
|
|
@@ -0,0 +1,117 @@
|
|
|
1
|
+
require "json"
|
|
2
|
+
|
|
3
|
+
module Silas
|
|
4
|
+
module Api
|
|
5
|
+
module V1
|
|
6
|
+
# Server-sent events at ROW granularity: turn / completed-step /
|
|
7
|
+
# invocation changes polled from the durable rows — the same source of
|
|
8
|
+
# truth as everything else. Deliberately NOT per-token: deltas are
|
|
9
|
+
# emitted in the worker process and the gem requires no cross-process
|
|
10
|
+
# bus; per-token streaming is the Turbo/browser feature.
|
|
11
|
+
#
|
|
12
|
+
# Delivery is at-least-once with Last-Event-ID resume: event ids are
|
|
13
|
+
# epoch-millisecond watermarks, each poll re-reads a 1ms overlap, and
|
|
14
|
+
# clients dedup by (event type, id, status). `?poll=1` emits the backlog
|
|
15
|
+
# since Last-Event-ID and closes — curl-able, and how the request specs
|
|
16
|
+
# exercise the cursor without holding a live stream.
|
|
17
|
+
#
|
|
18
|
+
# Live actions run on their own thread and hold a Puma thread for the
|
|
19
|
+
# stream's lifetime — inherent to SSE. The AR connection is NOT held:
|
|
20
|
+
# each poll checks one out and returns it before sleeping, and the
|
|
21
|
+
# stream closes itself after api_stream_max_duration (clients reconnect
|
|
22
|
+
# with Last-Event-ID).
|
|
23
|
+
class StreamsController < Silas::Api::BaseController
|
|
24
|
+
include ActionController::Live
|
|
25
|
+
include Silas::Api::Serialization
|
|
26
|
+
|
|
27
|
+
HEARTBEAT_EVERY = 15 # seconds; comment-frames keep proxies alive
|
|
28
|
+
|
|
29
|
+
def show
|
|
30
|
+
session_id = with_connection { Silas::Session.find(params[:id]).id }
|
|
31
|
+
|
|
32
|
+
response.headers["Content-Type"] = "text/event-stream"
|
|
33
|
+
response.headers["Cache-Control"] = "no-cache"
|
|
34
|
+
response.headers["X-Accel-Buffering"] = "no" # nginx: never buffer SSE
|
|
35
|
+
# Whatever the router/auth checked out must not stay pinned for the
|
|
36
|
+
# stream's lifetime.
|
|
37
|
+
ActiveRecord::Base.connection_handler.clear_active_connections!
|
|
38
|
+
|
|
39
|
+
watermark = initial_watermark
|
|
40
|
+
deadline = clock + Silas.config.api_stream_max_duration
|
|
41
|
+
last_beat = clock
|
|
42
|
+
|
|
43
|
+
loop do
|
|
44
|
+
events = with_connection { collect_changes(session_id, watermark) }
|
|
45
|
+
events.each do |type, at, payload|
|
|
46
|
+
write_event(type, payload, watermark_ms(at))
|
|
47
|
+
watermark = at if at > watermark
|
|
48
|
+
end
|
|
49
|
+
|
|
50
|
+
break if params[:poll].present? # backlog served; close
|
|
51
|
+
|
|
52
|
+
if clock >= deadline
|
|
53
|
+
write_event("timeout", { reconnect: true }, watermark_ms(watermark))
|
|
54
|
+
break
|
|
55
|
+
end
|
|
56
|
+
|
|
57
|
+
if clock - last_beat >= HEARTBEAT_EVERY
|
|
58
|
+
response.stream.write(": hb\n\n")
|
|
59
|
+
last_beat = clock
|
|
60
|
+
end
|
|
61
|
+
|
|
62
|
+
sleep Silas.config.api_stream_poll_interval
|
|
63
|
+
end
|
|
64
|
+
rescue IOError, ActionController::Live::ClientDisconnected
|
|
65
|
+
# client went away — the normal end of an SSE stream
|
|
66
|
+
ensure
|
|
67
|
+
response.stream.close rescue nil
|
|
68
|
+
end
|
|
69
|
+
|
|
70
|
+
private
|
|
71
|
+
|
|
72
|
+
# Everything with updated_at at-or-after the watermark (1ms overlap —
|
|
73
|
+
# at-least-once beats a gap). Steps only when completed: their text is
|
|
74
|
+
# the payload, and incomplete steps have none.
|
|
75
|
+
def collect_changes(session_id, watermark)
|
|
76
|
+
overlap = watermark - 0.001
|
|
77
|
+
events = []
|
|
78
|
+
|
|
79
|
+
Silas::Turn.where(session_id: session_id).where(updated_at: overlap..)
|
|
80
|
+
.order(:updated_at, :id).each do |turn|
|
|
81
|
+
events << [ "turn", turn.updated_at, turn_json(turn) ]
|
|
82
|
+
end
|
|
83
|
+
Silas::Step.joins(:turn).where(silas_turns: { session_id: session_id })
|
|
84
|
+
.where(status: "completed").where(updated_at: overlap..)
|
|
85
|
+
.order(:updated_at, :id).each do |step|
|
|
86
|
+
events << [ "step", step.updated_at, step_json(step) ]
|
|
87
|
+
end
|
|
88
|
+
Silas::ToolInvocation.joins(:turn).where(silas_turns: { session_id: session_id })
|
|
89
|
+
.where(updated_at: overlap..)
|
|
90
|
+
.order(:updated_at, :id).each do |invocation|
|
|
91
|
+
events << [ "invocation", invocation.updated_at, invocation_json(invocation) ]
|
|
92
|
+
end
|
|
93
|
+
|
|
94
|
+
events.sort_by { |(_type, at, _payload)| at }
|
|
95
|
+
end
|
|
96
|
+
|
|
97
|
+
def write_event(type, payload, id_ms)
|
|
98
|
+
response.stream.write("id: #{id_ms}\nevent: #{type}\ndata: #{JSON.generate(payload)}\n\n")
|
|
99
|
+
end
|
|
100
|
+
|
|
101
|
+
# Last-Event-ID (header or param) in epoch ms; absent -> "from now"
|
|
102
|
+
# (GET /sessions/:id carries current state; the stream carries what
|
|
103
|
+
# changes next). Last-Event-ID: 0 replays the whole session.
|
|
104
|
+
def initial_watermark
|
|
105
|
+
raw = request.headers["Last-Event-ID"].presence || params[:last_event_id].presence
|
|
106
|
+
raw ? Time.zone.at(raw.to_i / 1000.0) : Time.zone.now
|
|
107
|
+
end
|
|
108
|
+
|
|
109
|
+
def watermark_ms(time) = (time.to_f * 1000).round
|
|
110
|
+
|
|
111
|
+
def with_connection(&) = ActiveRecord::Base.connection_pool.with_connection(&)
|
|
112
|
+
|
|
113
|
+
def clock = Process.clock_gettime(Process::CLOCK_MONOTONIC)
|
|
114
|
+
end
|
|
115
|
+
end
|
|
116
|
+
end
|
|
117
|
+
end
|
|
@@ -0,0 +1,32 @@
|
|
|
1
|
+
module Silas
|
|
2
|
+
module Api
|
|
3
|
+
module V1
|
|
4
|
+
class TurnsController < Silas::Api::BaseController
|
|
5
|
+
include Silas::Api::Serialization
|
|
6
|
+
|
|
7
|
+
# POST /silas/api/v1/sessions/:session_id/turns { input: }
|
|
8
|
+
# 409 when a turn is already active — an API must surface it, not
|
|
9
|
+
# swallow it the way the webhook channels do.
|
|
10
|
+
def create
|
|
11
|
+
session = Silas::Session.find(params[:session_id])
|
|
12
|
+
input = params[:input].to_s.strip
|
|
13
|
+
return render json: { error: "input is required" }, status: :unprocessable_entity if input.empty?
|
|
14
|
+
|
|
15
|
+
turn = session.continue(input: input)
|
|
16
|
+
render json: turn_json(turn), status: :created
|
|
17
|
+
rescue Silas::TurnInProgressError => e
|
|
18
|
+
render json: { error: e.message }, status: :conflict
|
|
19
|
+
end
|
|
20
|
+
|
|
21
|
+
# POST /silas/api/v1/turns/:id/cancel
|
|
22
|
+
# Running turns are flagged (honored at the next step boundary);
|
|
23
|
+
# queued/parked turns cancel immediately — cancel: reflects which.
|
|
24
|
+
def cancel
|
|
25
|
+
turn = Silas::Turn.find(params[:id])
|
|
26
|
+
outcome = turn.cancel!(reason: "canceled via api by #{current_actor}")
|
|
27
|
+
render json: turn_json(turn.reload).merge(cancel: outcome)
|
|
28
|
+
end
|
|
29
|
+
end
|
|
30
|
+
end
|
|
31
|
+
end
|
|
32
|
+
end
|
|
@@ -3,18 +3,31 @@ module Silas
|
|
|
3
3
|
class SessionsController < BaseController
|
|
4
4
|
before_action :authenticate_write!, only: :create
|
|
5
5
|
|
|
6
|
+
PER_PAGE = 50
|
|
7
|
+
|
|
6
8
|
def index
|
|
7
|
-
|
|
8
|
-
|
|
9
|
+
# Keyset pagination on id (?before=<id> pages older), newest first.
|
|
10
|
+
scope = Silas::Session.order(id: :desc)
|
|
11
|
+
scope = scope.where(agent_name: params[:agent]) if params[:agent].present?
|
|
9
12
|
if params[:pending].present? # the "N awaiting approval" badge drills into this
|
|
10
13
|
# Subquery, not joins+distinct: DISTINCT over silas_sessions.* trips
|
|
11
14
|
# on the json metadata column (PG json has no equality operator).
|
|
12
|
-
|
|
15
|
+
scope = scope.where(
|
|
13
16
|
id: Silas::Turn.joins(:tool_invocations)
|
|
14
17
|
.where(silas_tool_invocations: { approval_state: "required" })
|
|
15
18
|
.select(:session_id)
|
|
16
19
|
)
|
|
17
20
|
end
|
|
21
|
+
scope = scope.where(id: ...params[:before].to_i) if params[:before].present?
|
|
22
|
+
|
|
23
|
+
# One query for the rows + turns, one for the pending counts — the
|
|
24
|
+
# per-row active_turn/turns.last/counts pattern was ~4 queries a card.
|
|
25
|
+
@sessions = scope.limit(PER_PAGE).includes(:turns).to_a
|
|
26
|
+
@next_before = @sessions.last&.id if @sessions.size == PER_PAGE
|
|
27
|
+
@pending_counts = Silas::ToolInvocation.joins(:turn)
|
|
28
|
+
.where(approval_state: "required",
|
|
29
|
+
silas_turns: { session_id: @sessions.map(&:id) })
|
|
30
|
+
.group("silas_turns.session_id").count
|
|
18
31
|
@agent_names = Silas::Session.distinct.pluck(:agent_name).sort
|
|
19
32
|
@pending_total = Silas::ToolInvocation.where(approval_state: "required").count
|
|
20
33
|
end
|
|
@@ -27,6 +27,25 @@ module Silas
|
|
|
27
27
|
notice = outcome == :cancel_requested ? "Cancel requested — honored at the next step boundary." : "Turn canceled."
|
|
28
28
|
redirect_to inbox_session_path(turn.session_id), notice: notice
|
|
29
29
|
end
|
|
30
|
+
|
|
31
|
+
# Top up a budget-parked turn and resume it — the CHANGELOG promised
|
|
32
|
+
# this card in 0.1.5; completed steps replay from rows, same as approvals.
|
|
33
|
+
def raise_budget
|
|
34
|
+
turn = Silas::Turn.find(params[:id])
|
|
35
|
+
unless turn.budget_parked?
|
|
36
|
+
return redirect_to inbox_session_path(turn.session_id),
|
|
37
|
+
alert: "Turn ##{turn.index} is not budget-parked."
|
|
38
|
+
end
|
|
39
|
+
|
|
40
|
+
reason = turn.failure_reason.to_s
|
|
41
|
+
value = params[:value].to_s
|
|
42
|
+
numeric = reason == "max_cost" ? value.to_f : value.to_i
|
|
43
|
+
|
|
44
|
+
turn.raise_budget!(**{ reason.to_sym => numeric })
|
|
45
|
+
redirect_to inbox_session_path(turn.session_id), notice: "#{reason} raised — resuming."
|
|
46
|
+
rescue Silas::Error, ArgumentError => e
|
|
47
|
+
redirect_to inbox_session_path(turn.session_id), alert: e.message
|
|
48
|
+
end
|
|
30
49
|
end
|
|
31
50
|
end
|
|
32
51
|
end
|
|
@@ -18,6 +18,11 @@ module Silas
|
|
|
18
18
|
Array(step.response_blocks).select { |b| b["type"] == "text" }.map { |b| b["text"] }.join.presence
|
|
19
19
|
end
|
|
20
20
|
|
|
21
|
+
# The final_answer payload, when the agent declared a schema.
|
|
22
|
+
def step_structured(step)
|
|
23
|
+
Array(step.response_blocks).reverse.find { |b| b["type"] == "structured" }&.dig("data")
|
|
24
|
+
end
|
|
25
|
+
|
|
21
26
|
def pretty_args(hash)
|
|
22
27
|
JSON.pretty_generate(hash || {})
|
|
23
28
|
end
|
data/app/models/silas/session.rb
CHANGED
|
@@ -5,6 +5,12 @@ module Silas
|
|
|
5
5
|
has_many :turns, -> { order(:index) }, class_name: "Silas::Turn", foreign_key: :session_id,
|
|
6
6
|
inverse_of: :session, dependent: :destroy
|
|
7
7
|
|
|
8
|
+
# Delegations and handoffs stamp parent_session_id; without associations
|
|
9
|
+
# they surfaced as unexplained orphans.
|
|
10
|
+
belongs_to :parent_session, class_name: "Silas::Session", optional: true
|
|
11
|
+
has_many :child_sessions, class_name: "Silas::Session",
|
|
12
|
+
foreign_key: :parent_session_id, inverse_of: :parent_session
|
|
13
|
+
|
|
8
14
|
validates :status, inclusion: { in: STATUSES }
|
|
9
15
|
validates :agent_name, presence: true
|
|
10
16
|
|
data/app/models/silas/turn.rb
CHANGED
|
@@ -84,6 +84,15 @@ module Silas
|
|
|
84
84
|
Array(step.response_blocks).select { |b| b["type"] == "text" }.map { |b| b["text"] }.join
|
|
85
85
|
end
|
|
86
86
|
|
|
87
|
+
# The structured payload when the agent declares a final_answer schema in
|
|
88
|
+
# agent.yml (nil otherwise) — the parsed Hash, not a string to re-parse.
|
|
89
|
+
def answer_data
|
|
90
|
+
step = steps.where(status: "completed").order(:index).last
|
|
91
|
+
return nil unless step
|
|
92
|
+
|
|
93
|
+
Array(step.response_blocks).reverse.find { |b| b["type"] == "structured" }&.dig("data")
|
|
94
|
+
end
|
|
95
|
+
|
|
87
96
|
# Outbound: when a channel-bound turn reaches an answer, deliver it off-loop.
|
|
88
97
|
after_update_commit :notify_channel_answer, if: :should_notify_answer?
|
|
89
98
|
|
|
@@ -92,6 +92,12 @@
|
|
|
92
92
|
.agent-filter .chip { font-size: 12px; padding: 3px 10px; border: 1px solid #d9dce1;
|
|
93
93
|
border-radius: 999px; text-decoration: none; color: inherit; }
|
|
94
94
|
.agent-filter .chip-on { background: #16181d; color: #fff; border-color: #16181d; }
|
|
95
|
+
.pager { text-align: center; margin: 16px 0; }
|
|
96
|
+
.pager .chip { font-size: 13px; padding: 6px 14px; border: 1px solid var(--line);
|
|
97
|
+
border-radius: 999px; text-decoration: none; color: inherit; }
|
|
98
|
+
.topup-form { display: flex; gap: 8px; margin-top: 8px; }
|
|
99
|
+
.topup-input { flex: 1; border: 1px solid var(--line); border-radius: 8px; padding: 8px 10px;
|
|
100
|
+
font: inherit; background: var(--panel); color: var(--ink); }
|
|
95
101
|
</style>
|
|
96
102
|
</head>
|
|
97
103
|
<body>
|
|
@@ -37,15 +37,23 @@
|
|
|
37
37
|
<%= link_to inbox_session_path(session), class: "card session-row" do %>
|
|
38
38
|
<div class="row-top">
|
|
39
39
|
<span class="name"><%= session.agent_name %></span>
|
|
40
|
-
|
|
41
|
-
<%
|
|
40
|
+
<%# Loaded association — no per-row queries. %>
|
|
41
|
+
<% if (turn = session.turns.detect(&:active?) || session.turns.last) %><%= status_pill(turn.status) %><% end %>
|
|
42
|
+
<% pending = @pending_counts.fetch(session.id, 0) %>
|
|
42
43
|
<% if pending.positive? %><span class="badge"><%= pending %> to approve</span><% end %>
|
|
43
44
|
</div>
|
|
44
45
|
<div class="muted">
|
|
45
|
-
<%= pluralize(session.turns.
|
|
46
|
+
<%= pluralize(session.turns.size, "turn") %> ·
|
|
46
47
|
<%= session.channel.presence || "direct" %> ·
|
|
47
48
|
<%= relative_time(session.updated_at) %>
|
|
48
49
|
</div>
|
|
49
50
|
<% end %>
|
|
50
51
|
<% end %>
|
|
52
|
+
<% if @next_before %>
|
|
53
|
+
<div class="pager">
|
|
54
|
+
<%= link_to "Older sessions →",
|
|
55
|
+
inbox_sessions_path(before: @next_before, agent: params[:agent].presence, pending: params[:pending].presence),
|
|
56
|
+
class: "chip" %>
|
|
57
|
+
</div>
|
|
58
|
+
<% end %>
|
|
51
59
|
<% end %>
|
|
@@ -7,6 +7,9 @@
|
|
|
7
7
|
<% else %>
|
|
8
8
|
<div class="step-text step-live" id="silas-step-<%= step.id %>-text"></div>
|
|
9
9
|
<% end %>
|
|
10
|
+
<% if (data = step_structured(step)) %>
|
|
11
|
+
<pre class="args"><%= pretty_args(data) %></pre>
|
|
12
|
+
<% end %>
|
|
10
13
|
<div id="silas-step-<%= step.id %>-tools">
|
|
11
14
|
<%= render partial: "silas/inbox/invocations/invocation", collection: step.tool_invocations, as: :invocation %>
|
|
12
15
|
</div>
|
|
@@ -9,6 +9,17 @@
|
|
|
9
9
|
<span class="muted">cancel requested…</span>
|
|
10
10
|
<% end %>
|
|
11
11
|
</div>
|
|
12
|
-
<% if turn.
|
|
12
|
+
<% if turn.budget_parked? %>
|
|
13
|
+
<div class="approval">
|
|
14
|
+
<h3>Budget reached — <%= turn.failure_reason %></h3>
|
|
15
|
+
<div class="muted">Parked at zero compute. Raise the limit and the turn resumes
|
|
16
|
+
by replaying completed work from rows — no re-calls, no re-effects.</div>
|
|
17
|
+
<%= form_with url: raise_budget_inbox_turn_path(turn), method: :post, class: "topup-form" do %>
|
|
18
|
+
<input name="value" type="number" step="any" min="0" required class="topup-input"
|
|
19
|
+
placeholder="<%= turn.failure_reason == "max_cost" ? "new limit in dollars, e.g. 2.50" : "new limit, e.g. 200000" %>">
|
|
20
|
+
<button class="btn approve">Raise & resume</button>
|
|
21
|
+
<% end %>
|
|
22
|
+
</div>
|
|
23
|
+
<% elsif turn.failure_reason.present? %>
|
|
13
24
|
<div class="muted">failed: <%= turn.failure_reason %></div>
|
|
14
25
|
<% end %>
|